Tcl Source Code

*/config.cache
*/config.log
*/config.status
*/tclConfig.sh
*/tclsh*
*/tcltest*
*/versions.vc












unix/autoMkindex.tcl
unix/dltest.marker
unix/tcl.pc
unix/tclIndex
unix/pkgs/*
win/pkgs/*
win/tcl.hpj

This is makeindex, version 2.14 [02-Oct-2002] (kpathsea + Thai support).
Scanning input file bn.idx....done (79 entries accepted, 0 rejected).
Sorting entries....done (511 comparisons).
Generating output file bn.ind....done (82 lines written, 0 warnings).
Output written in bn.ind.
Transcript written in bn.ilg.

\begin{theindex}

  \item mp\_add, \hyperpage{29}
  \item mp\_add\_d, \hyperpage{52}
  \item mp\_and, \hyperpage{29}
  \item mp\_clear, \hyperpage{11}
  \item mp\_clear\_multi, \hyperpage{12}
  \item mp\_cmp, \hyperpage{24}
  \item mp\_cmp\_d, \hyperpage{25}
  \item mp\_cmp\_mag, \hyperpage{23}
  \item mp\_div, \hyperpage{30}
  \item mp\_div\_2, \hyperpage{26}
  \item mp\_div\_2d, \hyperpage{28}
  \item mp\_div\_d, \hyperpage{52}
  \item mp\_dr\_reduce, \hyperpage{40}
  \item mp\_dr\_setup, \hyperpage{40}
  \item MP\_EQ, \hyperpage{22}
  \item mp\_error\_to\_string, \hyperpage{10}
  \item mp\_expt\_d, \hyperpage{43}
  \item mp\_exptmod, \hyperpage{43}
  \item mp\_exteuclid, \hyperpage{51}
  \item mp\_gcd, \hyperpage{51}
  \item mp\_get\_int, \hyperpage{20}
  \item mp\_grow, \hyperpage{16}
  \item MP\_GT, \hyperpage{22}
  \item mp\_init, \hyperpage{11}
  \item mp\_init\_copy, \hyperpage{13}
  \item mp\_init\_multi, \hyperpage{12}
  \item mp\_init\_set, \hyperpage{21}
  \item mp\_init\_set\_int, \hyperpage{21}
  \item mp\_init\_size, \hyperpage{14}
  \item mp\_int, \hyperpage{10}
  \item mp\_invmod, \hyperpage{52}
  \item mp\_jacobi, \hyperpage{52}
  \item mp\_lcm, \hyperpage{51}
  \item mp\_lshd, \hyperpage{28}
  \item MP\_LT, \hyperpage{22}
  \item MP\_MEM, \hyperpage{9}
  \item mp\_mod, \hyperpage{35}
  \item mp\_mod\_d, \hyperpage{52}
  \item mp\_montgomery\_calc\_normalization, \hyperpage{38}
  \item mp\_montgomery\_reduce, \hyperpage{37}
  \item mp\_montgomery\_setup, \hyperpage{37}
  \item mp\_mul, \hyperpage{31}
  \item mp\_mul\_2, \hyperpage{26}
  \item mp\_mul\_2d, \hyperpage{28}
  \item mp\_mul\_d, \hyperpage{52}
  \item mp\_n\_root, \hyperpage{44}
  \item mp\_neg, \hyperpage{29}
  \item MP\_NO, \hyperpage{9}
  \item MP\_OKAY, \hyperpage{9}
  \item mp\_or, \hyperpage{29}
  \item mp\_prime\_fermat, \hyperpage{45}
  \item mp\_prime\_is\_divisible, \hyperpage{45}
  \item mp\_prime\_is\_prime, \hyperpage{46}
  \item mp\_prime\_miller\_rabin, \hyperpage{45}
  \item mp\_prime\_next\_prime, \hyperpage{46}
  \item mp\_prime\_rabin\_miller\_trials, \hyperpage{46}
  \item mp\_prime\_random, \hyperpage{47}
  \item mp\_prime\_random\_ex, \hyperpage{47}
  \item mp\_radix\_size, \hyperpage{49}
  \item mp\_read\_radix, \hyperpage{49}
  \item mp\_read\_unsigned\_bin, \hyperpage{50}
  \item mp\_reduce, \hyperpage{36}
  \item mp\_reduce\_2k, \hyperpage{41}
  \item mp\_reduce\_2k\_setup, \hyperpage{41}
  \item mp\_reduce\_setup, \hyperpage{36}
  \item mp\_rshd, \hyperpage{28}
  \item mp\_set, \hyperpage{19}
  \item mp\_set\_int, \hyperpage{20}
  \item mp\_shrink, \hyperpage{15}
  \item mp\_sqr, \hyperpage{33}
  \item mp\_sub, \hyperpage{29}
  \item mp\_sub\_d, \hyperpage{52}
  \item mp\_to\_unsigned\_bin, \hyperpage{50}
  \item mp\_toradix, \hyperpage{49}
  \item mp\_unsigned\_bin\_size, \hyperpage{50}
  \item MP\_VAL, \hyperpage{9}
  \item mp\_xor, \hyperpage{29}
  \item MP\_YES, \hyperpage{9}

\end{theindex}

#include <time.h>

#ifdef IOWNANATHLON
#include <unistd.h>
#define SLEEP sleep(4)
#else
#define SLEEP
#endif

#include "tommath.h"

void ndraw(mp_int * a, char *name)
{
   char buf[16000];

   printf("%s: ", name);
   mp_toradix(a, buf, 10);
   printf("%s\n", buf);
}

static void draw(mp_int * a)
{
   ndraw(a, "");
}


unsigned long lfsr = 0xAAAAAAAAUL;

int lbit(void)
{
   if (lfsr & 0x80000000UL) {
      lfsr = ((lfsr << 1) ^ 0x8000001BUL) & 0xFFFFFFFFUL;
      return 1;
   } else {
      lfsr <<= 1;
      return 0;
   }
}

int myrng(unsigned char *dst, int len, void *dat)
{
   int x;

   for (x = 0; x < len; x++)
      dst[x] = rand() & 0xFF;
   return len;
}



char cmd[4096], buf[4096];
int main(void)
{
   mp_int a, b, c, d, e, f;
   unsigned long expt_n, add_n, sub_n, mul_n, div_n, sqr_n, mul2d_n, div2d_n,
      gcd_n, lcm_n, inv_n, div2_n, mul2_n, add_d_n, sub_d_n, t;
   unsigned rr;
   int i, n, err, cnt, ix, old_kara_m, old_kara_s;
   mp_digit mp;


   mp_init(&a);
   mp_init(&b);
   mp_init(&c);
   mp_init(&d);
   mp_init(&e);
   mp_init(&f);

   srand(time(NULL));

#if 0
   // test montgomery 
   printf("Testing montgomery...\n");
   for (i = 1; i < 10; i++) {
      printf("Testing digit size: %d\n", i);
      for (n = 0; n < 1000; n++) {
         mp_rand(&a, i);
         a.dp[0] |= 1;

         // let's see if R is right
         mp_montgomery_calc_normalization(&b, &a);
         mp_montgomery_setup(&a, &mp);

         // now test a random reduction 
         for (ix = 0; ix < 100; ix++) {
             mp_rand(&c, 1 + abs(rand()) % (2*i));
             mp_copy(&c, &d);
             mp_copy(&c, &e);

             mp_mod(&d, &a, &d);
             mp_montgomery_reduce(&c, &a, mp);
             mp_mulmod(&c, &b, &a, &c);

             if (mp_cmp(&c, &d) != MP_EQ) { 
printf("d = e mod a, c = e MOD a\n");
mp_todecimal(&a, buf); printf("a = %s\n", buf);
mp_todecimal(&e, buf); printf("e = %s\n", buf);
mp_todecimal(&d, buf); printf("d = %s\n", buf);
mp_todecimal(&c, buf); printf("c = %s\n", buf);
printf("compare no compare!\n"); exit(EXIT_FAILURE); }
         }
      }
   }
   printf("done\n");

   // test mp_get_int
   printf("Testing: mp_get_int\n");
   for (i = 0; i < 1000; ++i) {
      t = ((unsigned long) rand() * rand() + 1) & 0xFFFFFFFF;
      mp_set_int(&a, t);
      if (t != mp_get_int(&a)) {
	 printf("mp_get_int() bad result!\n");
	 return 1;
      }
   }
   mp_set_int(&a, 0);
   if (mp_get_int(&a) != 0) {
      printf("mp_get_int() bad result!\n");
      return 1;
   }
   mp_set_int(&a, 0xffffffff);
   if (mp_get_int(&a) != 0xffffffff) {
      printf("mp_get_int() bad result!\n");
      return 1;
   }
   // test mp_sqrt
   printf("Testing: mp_sqrt\n");
   for (i = 0; i < 1000; ++i) {
      printf("%6d\r", i);
      fflush(stdout);
      n = (rand() & 15) + 1;
      mp_rand(&a, n);
      if (mp_sqrt(&a, &b) != MP_OKAY) {
	 printf("mp_sqrt() error!\n");
	 return 1;
      }
      mp_n_root(&a, 2, &a);
      if (mp_cmp_mag(&b, &a) != MP_EQ) {
	 printf("mp_sqrt() bad result!\n");
	 return 1;
      }
   }

   printf("\nTesting: mp_is_square\n");
   for (i = 0; i < 1000; ++i) {
      printf("%6d\r", i);
      fflush(stdout);

      /* test mp_is_square false negatives */
      n = (rand() & 7) + 1;
      mp_rand(&a, n);
      mp_sqr(&a, &a);
      if (mp_is_square(&a, &n) != MP_OKAY) {
	 printf("fn:mp_is_square() error!\n");
	 return 1;
      }
      if (n == 0) {
	 printf("fn:mp_is_square() bad result!\n");
	 return 1;
      }

      /* test for false positives */
      mp_add_d(&a, 1, &a);
      if (mp_is_square(&a, &n) != MP_OKAY) {
	 printf("fp:mp_is_square() error!\n");
	 return 1;
      }
      if (n == 1) {
	 printf("fp:mp_is_square() bad result!\n");
	 return 1;
      }

   }
   printf("\n\n");

   /* test for size */
   for (ix = 10; ix < 128; ix++) {
      printf("Testing (not safe-prime): %9d bits    \r", ix);
      fflush(stdout);
      err =
	 mp_prime_random_ex(&a, 8, ix,
			    (rand() & 1) ? LTM_PRIME_2MSB_OFF :
			    LTM_PRIME_2MSB_ON, myrng, NULL);
      if (err != MP_OKAY) {
	 printf("failed with err code %d\n", err);
	 return EXIT_FAILURE;
      }
      if (mp_count_bits(&a) != ix) {
	 printf("Prime is %d not %d bits!!!\n", mp_count_bits(&a), ix);
	 return EXIT_FAILURE;
      }
   }

   for (ix = 16; ix < 128; ix++) {
      printf("Testing (   safe-prime): %9d bits    \r", ix);
      fflush(stdout);
      err =
	 mp_prime_random_ex(&a, 8, ix,
			    ((rand() & 1) ? LTM_PRIME_2MSB_OFF :
			     LTM_PRIME_2MSB_ON) | LTM_PRIME_SAFE, myrng,
			    NULL);
      if (err != MP_OKAY) {
	 printf("failed with err code %d\n", err);
	 return EXIT_FAILURE;
      }
      if (mp_count_bits(&a) != ix) {
	 printf("Prime is %d not %d bits!!!\n", mp_count_bits(&a), ix);
	 return EXIT_FAILURE;
      }
      /* let's see if it's really a safe prime */
      mp_sub_d(&a, 1, &a);
      mp_div_2(&a, &a);
      mp_prime_is_prime(&a, 8, &cnt);
      if (cnt != MP_YES) {
	 printf("sub is not prime!\n");
	 return EXIT_FAILURE;
      }
   }

   printf("\n\n");

   mp_read_radix(&a, "123456", 10);
   mp_toradix_n(&a, buf, 10, 3);
   printf("a == %s\n", buf);
   mp_toradix_n(&a, buf, 10, 4);
   printf("a == %s\n", buf);
   mp_toradix_n(&a, buf, 10, 30);
   printf("a == %s\n", buf);


#if 0
   for (;;) {
      fgets(buf, sizeof(buf), stdin);
      mp_read_radix(&a, buf, 10);
      mp_prime_next_prime(&a, 5, 1);
      mp_toradix(&a, buf, 10);
      printf("%s, %lu\n", buf, a.dp[0] & 3);
   }
#endif

   /* test mp_cnt_lsb */
   printf("testing mp_cnt_lsb...\n");
   mp_set(&a, 1);
   for (ix = 0; ix < 1024; ix++) {
      if (mp_cnt_lsb(&a) != ix) {
	 printf("Failed at %d, %d\n", ix, mp_cnt_lsb(&a));
	 return 0;
      }
      mp_mul_2(&a, &a);
   }

/* test mp_reduce_2k */
   printf("Testing mp_reduce_2k...\n");
   for (cnt = 3; cnt <= 128; ++cnt) {
      mp_digit tmp;

      mp_2expt(&a, cnt);
      mp_sub_d(&a, 2, &a);	/* a = 2**cnt - 2 */


      printf("\nTesting %4d bits", cnt);
      printf("(%d)", mp_reduce_is_2k(&a));
      mp_reduce_2k_setup(&a, &tmp);
      printf("(%d)", tmp);
      for (ix = 0; ix < 1000; ix++) {
	 if (!(ix & 127)) {
	    printf(".");
	    fflush(stdout);
	 }
	 mp_rand(&b, (cnt / DIGIT_BIT + 1) * 2);
	 mp_copy(&c, &b);
	 mp_mod(&c, &a, &c);
	 mp_reduce_2k(&b, &a, 2);
	 if (mp_cmp(&c, &b)) {
	    printf("FAILED\n");
	    exit(0);
	 }
      }
   }

/* test mp_div_3  */
   printf("Testing mp_div_3...\n");
   mp_set(&d, 3);
   for (cnt = 0; cnt < 10000;) {
      mp_digit r1, r2;

      if (!(++cnt & 127))
	 printf("%9d\r", cnt);
      mp_rand(&a, abs(rand()) % 128 + 1);
      mp_div(&a, &d, &b, &e);
      mp_div_3(&a, &c, &r2);

      if (mp_cmp(&b, &c) || mp_cmp_d(&e, r2)) {
	 printf("\n\nmp_div_3 => Failure\n");
      }
   }
   printf("\n\nPassed div_3 testing\n");

/* test the DR reduction */
   printf("testing mp_dr_reduce...\n");
   for (cnt = 2; cnt < 32; cnt++) {
      printf("%d digit modulus\n", cnt);
      mp_grow(&a, cnt);
      mp_zero(&a);
      for (ix = 1; ix < cnt; ix++) {
	 a.dp[ix] = MP_MASK;
      }
      a.used = cnt;
      a.dp[0] = 3;

      mp_rand(&b, cnt - 1);
      mp_copy(&b, &c);

      rr = 0;
      do {
	 if (!(rr & 127)) {
	    printf("%9lu\r", rr);
	    fflush(stdout);
	 }
	 mp_sqr(&b, &b);
	 mp_add_d(&b, 1, &b);
	 mp_copy(&b, &c);

	 mp_mod(&b, &a, &b);
	 mp_dr_reduce(&c, &a, (((mp_digit) 1) << DIGIT_BIT) - a.dp[0]);

	 if (mp_cmp(&b, &c) != MP_EQ) {
	    printf("Failed on trial %lu\n", rr);
	    exit(-1);

	 }
      } while (++rr < 500);
      printf("Passed DR test for %d digits\n", cnt);
   }

#endif

/* test the mp_reduce_2k_l code */
#if 0
#if 0
/* first load P with 2^1024 - 0x2A434 B9FDEC95 D8F9D550 FFFFFFFF FFFFFFFF */
   mp_2expt(&a, 1024);
   mp_read_radix(&b, "2A434B9FDEC95D8F9D550FFFFFFFFFFFFFFFF", 16);
   mp_sub(&a, &b, &a);
#elif 1
/*  p = 2^2048 - 0x1 00000000 00000000 00000000 00000000 4945DDBF 8EA2A91D 5776399B B83E188F  */
   mp_2expt(&a, 2048);
   mp_read_radix(&b,
		 "1000000000000000000000000000000004945DDBF8EA2A91D5776399BB83E188F",
		 16);
   mp_sub(&a, &b, &a);
#endif

   mp_todecimal(&a, buf);
   printf("p==%s\n", buf);
/* now mp_reduce_is_2k_l() should return */
   if (mp_reduce_is_2k_l(&a) != 1) {
      printf("mp_reduce_is_2k_l() return 0, should be 1\n");
      return EXIT_FAILURE;
   }
   mp_reduce_2k_setup_l(&a, &d);
   /* now do a million square+1 to see if it varies */
   mp_rand(&b, 64);
   mp_mod(&b, &a, &b);
   mp_copy(&b, &c);
   printf("testing mp_reduce_2k_l...");
   fflush(stdout);
   for (cnt = 0; cnt < (1UL << 20); cnt++) {
      mp_sqr(&b, &b);
      mp_add_d(&b, 1, &b);
      mp_reduce_2k_l(&b, &a, &d);
      mp_sqr(&c, &c);
      mp_add_d(&c, 1, &c);
      mp_mod(&c, &a, &c);
      if (mp_cmp(&b, &c) != MP_EQ) {
	 printf("mp_reduce_2k_l() failed at step %lu\n", cnt);
	 mp_tohex(&b, buf);
	 printf("b == %s\n", buf);
	 mp_tohex(&c, buf);
	 printf("c == %s\n", buf);
	 return EXIT_FAILURE;
      }
   }
   printf("...Passed\n");
#endif

   div2_n = mul2_n = inv_n = expt_n = lcm_n = gcd_n = add_n =
      sub_n = mul_n = div_n = sqr_n = mul2d_n = div2d_n = cnt = add_d_n =
      sub_d_n = 0;

   /* force KARA and TOOM to enable despite cutoffs */
   KARATSUBA_SQR_CUTOFF = KARATSUBA_MUL_CUTOFF = 8;
   TOOM_SQR_CUTOFF = TOOM_MUL_CUTOFF = 16;

   for (;;) {
      /* randomly clear and re-init one variable, this has the affect of triming the alloc space */
      switch (abs(rand()) % 7) {
      case 0:
	 mp_clear(&a);
	 mp_init(&a);
	 break;
      case 1:
	 mp_clear(&b);
	 mp_init(&b);
	 break;
      case 2:
	 mp_clear(&c);
	 mp_init(&c);
	 break;
      case 3:
	 mp_clear(&d);
	 mp_init(&d);
	 break;
      case 4:
	 mp_clear(&e);
	 mp_init(&e);
	 break;
      case 5:
	 mp_clear(&f);
	 mp_init(&f);
	 break;
      case 6:
	 break;			/* don't clear any */
      }


      printf
	 ("%4lu/%4lu/%4lu/%4lu/%4lu/%4lu/%4lu/%4lu/%4lu/%4lu/%4lu/%4lu/%4lu/%4lu/%4lu ",
	  add_n, sub_n, mul_n, div_n, sqr_n, mul2d_n, div2d_n, gcd_n, lcm_n,
	  expt_n, inv_n, div2_n, mul2_n, add_d_n, sub_d_n);
      fgets(cmd, 4095, stdin);
      cmd[strlen(cmd) - 1] = 0;
      printf("%s  ]\r", cmd);
      fflush(stdout);
      if (!strcmp(cmd, "mul2d")) {
	 ++mul2d_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 sscanf(buf, "%d", &rr);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);

	 mp_mul_2d(&a, rr, &a);
	 a.sign = b.sign;
	 if (mp_cmp(&a, &b) != MP_EQ) {
	    printf("mul2d failed, rr == %d\n", rr);
	    draw(&a);
	    draw(&b);
	    return 0;
	 }
      } else if (!strcmp(cmd, "div2d")) {
	 ++div2d_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 sscanf(buf, "%d", &rr);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);

	 mp_div_2d(&a, rr, &a, &e);
	 a.sign = b.sign;
	 if (a.used == b.used && a.used == 0) {
	    a.sign = b.sign = MP_ZPOS;
	 }
	 if (mp_cmp(&a, &b) != MP_EQ) {
	    printf("div2d failed, rr == %d\n", rr);
	    draw(&a);
	    draw(&b);
	    return 0;
	 }
      } else if (!strcmp(cmd, "add")) {
	 ++add_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&c, buf, 64);
	 mp_copy(&a, &d);
	 mp_add(&d, &b, &d);
	 if (mp_cmp(&c, &d) != MP_EQ) {
	    printf("add %lu failure!\n", add_n);
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    draw(&d);
	    return 0;
	 }

	 /* test the sign/unsigned storage functions */

	 rr = mp_signed_bin_size(&c);
	 mp_to_signed_bin(&c, (unsigned char *) cmd);
	 memset(cmd + rr, rand() & 255, sizeof(cmd) - rr);
	 mp_read_signed_bin(&d, (unsigned char *) cmd, rr);
	 if (mp_cmp(&c, &d) != MP_EQ) {
	    printf("mp_signed_bin failure!\n");
	    draw(&c);
	    draw(&d);
	    return 0;
	 }


	 rr = mp_unsigned_bin_size(&c);
	 mp_to_unsigned_bin(&c, (unsigned char *) cmd);
	 memset(cmd + rr, rand() & 255, sizeof(cmd) - rr);
	 mp_read_unsigned_bin(&d, (unsigned char *) cmd, rr);
	 if (mp_cmp_mag(&c, &d) != MP_EQ) {
	    printf("mp_unsigned_bin failure!\n");
	    draw(&c);
	    draw(&d);
	    return 0;
	 }

      } else if (!strcmp(cmd, "sub")) {
	 ++sub_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&c, buf, 64);
	 mp_copy(&a, &d);
	 mp_sub(&d, &b, &d);
	 if (mp_cmp(&c, &d) != MP_EQ) {
	    printf("sub %lu failure!\n", sub_n);
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    draw(&d);
	    return 0;
	 }
      } else if (!strcmp(cmd, "mul")) {
	 ++mul_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&c, buf, 64);
	 mp_copy(&a, &d);
	 mp_mul(&d, &b, &d);
	 if (mp_cmp(&c, &d) != MP_EQ) {
	    printf("mul %lu failure!\n", mul_n);
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    draw(&d);
	    return 0;
	 }
      } else if (!strcmp(cmd, "div")) {
	 ++div_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&c, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&d, buf, 64);

	 mp_div(&a, &b, &e, &f);
	 if (mp_cmp(&c, &e) != MP_EQ || mp_cmp(&d, &f) != MP_EQ) {
	    printf("div %lu %d, %d, failure!\n", div_n, mp_cmp(&c, &e),
		   mp_cmp(&d, &f));
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    draw(&d);
	    draw(&e);
	    draw(&f);
	    return 0;
	 }

      } else if (!strcmp(cmd, "sqr")) {
	 ++sqr_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 mp_copy(&a, &c);
	 mp_sqr(&c, &c);
	 if (mp_cmp(&b, &c) != MP_EQ) {
	    printf("sqr %lu failure!\n", sqr_n);
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    return 0;
	 }
      } else if (!strcmp(cmd, "gcd")) {
	 ++gcd_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&c, buf, 64);
	 mp_copy(&a, &d);
	 mp_gcd(&d, &b, &d);
	 d.sign = c.sign;
	 if (mp_cmp(&c, &d) != MP_EQ) {
	    printf("gcd %lu failure!\n", gcd_n);
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    draw(&d);
	    return 0;
	 }
      } else if (!strcmp(cmd, "lcm")) {
	 ++lcm_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&c, buf, 64);
	 mp_copy(&a, &d);
	 mp_lcm(&d, &b, &d);
	 d.sign = c.sign;
	 if (mp_cmp(&c, &d) != MP_EQ) {
	    printf("lcm %lu failure!\n", lcm_n);
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    draw(&d);
	    return 0;
	 }
      } else if (!strcmp(cmd, "expt")) {
	 ++expt_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&c, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&d, buf, 64);
	 mp_copy(&a, &e);
	 mp_exptmod(&e, &b, &c, &e);
	 if (mp_cmp(&d, &e) != MP_EQ) {
	    printf("expt %lu failure!\n", expt_n);
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    draw(&d);
	    draw(&e);
	    return 0;
	 }
      } else if (!strcmp(cmd, "invmod")) {
	 ++inv_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&c, buf, 64);
	 mp_invmod(&a, &b, &d);
	 mp_mulmod(&d, &a, &b, &e);
	 if (mp_cmp_d(&e, 1) != MP_EQ) {
	    printf("inv [wrong value from MPI?!] failure\n");
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    draw(&d);
	    mp_gcd(&a, &b, &e);
	    draw(&e);
	    return 0;
	 }

      } else if (!strcmp(cmd, "div2")) {
	 ++div2_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 mp_div_2(&a, &c);
	 if (mp_cmp(&c, &b) != MP_EQ) {
	    printf("div_2 %lu failure\n", div2_n);
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    return 0;
	 }
      } else if (!strcmp(cmd, "mul2")) {
	 ++mul2_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 mp_mul_2(&a, &c);
	 if (mp_cmp(&c, &b) != MP_EQ) {
	    printf("mul_2 %lu failure\n", mul2_n);
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    return 0;
	 }
      } else if (!strcmp(cmd, "add_d")) {
	 ++add_d_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 sscanf(buf, "%d", &ix);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 mp_add_d(&a, ix, &c);
	 if (mp_cmp(&b, &c) != MP_EQ) {
	    printf("add_d %lu failure\n", add_d_n);
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    printf("d == %d\n", ix);
	    return 0;
	 }
      } else if (!strcmp(cmd, "sub_d")) {
	 ++sub_d_n;
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&a, buf, 64);
	 fgets(buf, 4095, stdin);
	 sscanf(buf, "%d", &ix);
	 fgets(buf, 4095, stdin);
	 mp_read_radix(&b, buf, 64);
	 mp_sub_d(&a, ix, &c);
	 if (mp_cmp(&b, &c) != MP_EQ) {
	    printf("sub_d %lu failure\n", sub_d_n);
	    draw(&a);
	    draw(&b);
	    draw(&c);
	    printf("d == %d\n", ix);
	    return 0;
	 }
      }
   }
   return 0;
}

#include <tommath.h>
#include <time.h>

ulong64 _tt;

#ifdef IOWNANATHLON
#include <unistd.h>
#define SLEEP sleep(4)
#else
#define SLEEP
#endif


void ndraw(mp_int * a, char *name)
{
   char buf[4096];

   printf("%s: ", name);
   mp_toradix(a, buf, 64);
   printf("%s\n", buf);
}

static void draw(mp_int * a)
{
   ndraw(a, "");
}


unsigned long lfsr = 0xAAAAAAAAUL;

int lbit(void)
{
   if (lfsr & 0x80000000UL) {
      lfsr = ((lfsr << 1) ^ 0x8000001BUL) & 0xFFFFFFFFUL;
      return 1;
   } else {
      lfsr <<= 1;
      return 0;
   }
}

/* RDTSC from Scott Duplichan */
static ulong64 TIMFUNC(void)
{
#if defined __GNUC__
#if defined(__i386__) || defined(__x86_64__)
   unsigned long long a;
   __asm__ __volatile__("rdtsc\nmovl %%eax,%0\nmovl %%edx,4+%0\n"::
			"m"(a):"%eax", "%edx");
   return a;
#else /* gcc-IA64 version */
   unsigned long result;
   __asm__ __volatile__("mov %0=ar.itc":"=r"(result)::"memory");

   while (__builtin_expect((int) result == -1, 0))
      __asm__ __volatile__("mov %0=ar.itc":"=r"(result)::"memory");

   return result;
#endif

   // Microsoft and Intel Windows compilers
#elif defined _M_IX86
   __asm rdtsc
#elif defined _M_AMD64
   return __rdtsc();
#elif defined _M_IA64
#if defined __INTEL_COMPILER
#include <ia64intrin.h>
#endif
   return __getReg(3116);
#else
#error need rdtsc function for this build
#endif
}

#define DO(x) x; x;
//#define DO4(x) DO2(x); DO2(x);
//#define DO8(x) DO4(x); DO4(x);
//#define DO(x)  DO8(x); DO8(x);

int main(void)
{
   ulong64 tt, gg, CLK_PER_SEC;
   FILE *log, *logb, *logc, *logd;
   mp_int a, b, c, d, e, f;
   int n, cnt, ix, old_kara_m, old_kara_s;
   unsigned rr;

   mp_init(&a);
   mp_init(&b);
   mp_init(&c);
   mp_init(&d);
   mp_init(&e);
   mp_init(&f);

   srand(time(NULL));


   /* temp. turn off TOOM */
   TOOM_MUL_CUTOFF = TOOM_SQR_CUTOFF = 100000;

   CLK_PER_SEC = TIMFUNC();
   sleep(1);
   CLK_PER_SEC = TIMFUNC() - CLK_PER_SEC;

   printf("CLK_PER_SEC == %llu\n", CLK_PER_SEC);
   goto exptmod;
   log = fopen("logs/add.log", "w");
   for (cnt = 8; cnt <= 128; cnt += 8) {
      SLEEP;
      mp_rand(&a, cnt);
      mp_rand(&b, cnt);
      rr = 0;
      tt = -1;
      do {
	 gg = TIMFUNC();
	 DO(mp_add(&a, &b, &c));
	 gg = (TIMFUNC() - gg) >> 1;
	 if (tt > gg)
	    tt = gg;
      } while (++rr < 100000);
      printf("Adding\t\t%4d-bit => %9llu/sec, %9llu cycles\n",
	     mp_count_bits(&a), CLK_PER_SEC / tt, tt);
      fprintf(log, "%d %9llu\n", cnt * DIGIT_BIT, tt);
      fflush(log);
   }
   fclose(log);

   log = fopen("logs/sub.log", "w");
   for (cnt = 8; cnt <= 128; cnt += 8) {
      SLEEP;
      mp_rand(&a, cnt);
      mp_rand(&b, cnt);
      rr = 0;
      tt = -1;
      do {
	 gg = TIMFUNC();
	 DO(mp_sub(&a, &b, &c));
	 gg = (TIMFUNC() - gg) >> 1;
	 if (tt > gg)
	    tt = gg;
      } while (++rr < 100000);

      printf("Subtracting\t\t%4d-bit => %9llu/sec, %9llu cycles\n",
	     mp_count_bits(&a), CLK_PER_SEC / tt, tt);
      fprintf(log, "%d %9llu\n", cnt * DIGIT_BIT, tt);
      fflush(log);
   }
   fclose(log);

   /* do mult/square twice, first without karatsuba and second with */
 multtest:
   old_kara_m = KARATSUBA_MUL_CUTOFF;
   old_kara_s = KARATSUBA_SQR_CUTOFF;
   for (ix = 0; ix < 2; ix++) {
      printf("With%s Karatsuba\n", (ix == 0) ? "out" : "");

      KARATSUBA_MUL_CUTOFF = (ix == 0) ? 9999 : old_kara_m;
      KARATSUBA_SQR_CUTOFF = (ix == 0) ? 9999 : old_kara_s;

      log = fopen((ix == 0) ? "logs/mult.log" : "logs/mult_kara.log", "w");
      for (cnt = 4; cnt <= 10240 / DIGIT_BIT; cnt += 2) {
	 SLEEP;
	 mp_rand(&a, cnt);
	 mp_rand(&b, cnt);
	 rr = 0;
	 tt = -1;
	 do {
	    gg = TIMFUNC();
	    DO(mp_mul(&a, &b, &c));
	    gg = (TIMFUNC() - gg) >> 1;
	    if (tt > gg)
	       tt = gg;
	 } while (++rr < 100);
	 printf("Multiplying\t%4d-bit => %9llu/sec, %9llu cycles\n",
		mp_count_bits(&a), CLK_PER_SEC / tt, tt);
	 fprintf(log, "%d %9llu\n", mp_count_bits(&a), tt);
	 fflush(log);
      }
      fclose(log);

      log = fopen((ix == 0) ? "logs/sqr.log" : "logs/sqr_kara.log", "w");
      for (cnt = 4; cnt <= 10240 / DIGIT_BIT; cnt += 2) {
	 SLEEP;
	 mp_rand(&a, cnt);
	 rr = 0;
	 tt = -1;
	 do {
	    gg = TIMFUNC();
	    DO(mp_sqr(&a, &b));
	    gg = (TIMFUNC() - gg) >> 1;
	    if (tt > gg)
	       tt = gg;
	 } while (++rr < 100);
	 printf("Squaring\t%4d-bit => %9llu/sec, %9llu cycles\n",
		mp_count_bits(&a), CLK_PER_SEC / tt, tt);
	 fprintf(log, "%d %9llu\n", mp_count_bits(&a), tt);
	 fflush(log);
      }
      fclose(log);

   }
 exptmod:

   {
      char *primes[] = {
	 /* 2K large moduli */
	 "179769313486231590772930519078902473361797697894230657273430081157732675805500963132708477322407536021120113879871393357658789768814416622492847430639474124377767893424865485276302219601246094119453082952085005768838150682342462881473913110540827237163350510684586239334100047359817950870678242457666208137217",
	 "32317006071311007300714876688669951960444102669715484032130345427524655138867890893197201411522913463688717960921898019494119559150490921095088152386448283120630877367300996091750197750389652106796057638384067568276792218642619756161838094338476170470581645852036305042887575891541065808607552399123930385521914333389668342420684974786564569494856176035326322058077805659331026192708460314150258592864177116725943603718461857357598351152301645904403697613233287231227125684710820209725157101726931323469678542580656697935045997268352998638099733077152121140120031150424541696791951097529546801429027668869927491725169",
	 "1044388881413152506691752710716624382579964249047383780384233483283953907971557456848826811934997558340890106714439262837987573438185793607263236087851365277945956976543709998340361590134383718314428070011855946226376318839397712745672334684344586617496807908705803704071284048740118609114467977783598029006686938976881787785946905630190260940599579453432823469303026696443059025015972399867714215541693835559885291486318237914434496734087811872639496475100189041349008417061675093668333850551032972088269550769983616369411933015213796825837188091833656751221318492846368125550225998300412344784862595674492194617023806505913245610825731835380087608622102834270197698202313169017678006675195485079921636419370285375124784014907159135459982790513399611551794271106831134090584272884279791554849782954323534517065223269061394905987693002122963395687782878948440616007412945674919823050571642377154816321380631045902916136926708342856440730447899971901781465763473223850267253059899795996090799469201774624817718449867455659250178329070473119433165550807568221846571746373296884912819520317457002440926616910874148385078411929804522981857338977648103126085902995208257421855249796721729039744118165938433694823325696642096892124547425283",
	 /* 2K moduli mersenne primes */
	 "6864797660130609714981900799081393217269435300143305409394463459185543183397656052122559640661454554977296311391480858037121987999716643812574028291115057151",
	 "531137992816767098689588206552468627329593117727031923199444138200403559860852242739162502265229285668889329486246501015346579337652707239409519978766587351943831270835393219031728127",
	 "10407932194664399081925240327364085538615262247266704805319112350403608059673360298012239441732324184842421613954281007791383566248323464908139906605677320762924129509389220345773183349661583550472959420547689811211693677147548478866962501384438260291732348885311160828538416585028255604666224831890918801847068222203140521026698435488732958028878050869736186900714720710555703168729087",
	 "1475979915214180235084898622737381736312066145333169775147771216478570297878078949377407337049389289382748507531496480477281264838760259191814463365330269540496961201113430156902396093989090226259326935025281409614983499388222831448598601834318536230923772641390209490231836446899608210795482963763094236630945410832793769905399982457186322944729636418890623372171723742105636440368218459649632948538696905872650486914434637457507280441823676813517852099348660847172579408422316678097670224011990280170474894487426924742108823536808485072502240519452587542875349976558572670229633962575212637477897785501552646522609988869914013540483809865681250419497686697771007",
	 "259117086013202627776246767922441530941818887553125427303974923161874019266586362086201209516800483406550695241733194177441689509238807017410377709597512042313066624082916353517952311186154862265604547691127595848775610568757931191017711408826252153849035830401185072116424747461823031471398340229288074545677907941037288235820705892351068433882986888616658650280927692080339605869308790500409503709875902119018371991620994002568935113136548829739112656797303241986517250116412703509705427773477972349821676443446668383119322540099648994051790241624056519054483690809616061625743042361721863339415852426431208737266591962061753535748892894599629195183082621860853400937932839420261866586142503251450773096274235376822938649407127700846077124211823080804139298087057504713825264571448379371125032081826126566649084251699453951887789613650248405739378594599444335231188280123660406262468609212150349937584782292237144339628858485938215738821232393687046160677362909315071",
	 "190797007524439073807468042969529173669356994749940177394741882673528979787005053706368049835514900244303495954950709725762186311224148828811920216904542206960744666169364221195289538436845390250168663932838805192055137154390912666527533007309292687539092257043362517857366624699975402375462954490293259233303137330643531556539739921926201438606439020075174723029056838272505051571967594608350063404495977660656269020823960825567012344189908927956646011998057988548630107637380993519826582389781888135705408653045219655801758081251164080554609057468028203308718724654081055323215860189611391296030471108443146745671967766308925858547271507311563765171008318248647110097614890313562856541784154881743146033909602737947385055355960331855614540900081456378659068370317267696980001187750995491090350108417050917991562167972281070161305972518044872048331306383715094854938415738549894606070722584737978176686422134354526989443028353644037187375385397838259511833166416134323695660367676897722287918773420968982326089026150031515424165462111337527431154890666327374921446276833564519776797633875503548665093914556482031482248883127023777039667707976559857333357013727342079099064400455741830654320379350833236245819348824064783585692924881021978332974949906122664421376034687815350484991",

	 /* DR moduli */
	 "14059105607947488696282932836518693308967803494693489478439861164411992439598399594747002144074658928593502845729752797260025831423419686528151609940203368612079",
	 "101745825697019260773923519755878567461315282017759829107608914364075275235254395622580447400994175578963163918967182013639660669771108475957692810857098847138903161308502419410142185759152435680068435915159402496058513611411688900243039",
	 "736335108039604595805923406147184530889923370574768772191969612422073040099331944991573923112581267542507986451953227192970402893063850485730703075899286013451337291468249027691733891486704001513279827771740183629161065194874727962517148100775228363421083691764065477590823919364012917984605619526140821797602431",
	 "38564998830736521417281865696453025806593491967131023221754800625044118265468851210705360385717536794615180260494208076605798671660719333199513807806252394423283413430106003596332513246682903994829528690198205120921557533726473585751382193953592127439965050261476810842071573684505878854588706623484573925925903505747545471088867712185004135201289273405614415899438276535626346098904241020877974002916168099951885406379295536200413493190419727789712076165162175783",
	 "542189391331696172661670440619180536749994166415993334151601745392193484590296600979602378676624808129613777993466242203025054573692562689251250471628358318743978285860720148446448885701001277560572526947619392551574490839286458454994488665744991822837769918095117129546414124448777033941223565831420390846864429504774477949153794689948747680362212954278693335653935890352619041936727463717926744868338358149568368643403037768649616778526013610493696186055899318268339432671541328195724261329606699831016666359440874843103020666106568222401047720269951530296879490444224546654729111504346660859907296364097126834834235287147",
	 "1487259134814709264092032648525971038895865645148901180585340454985524155135260217788758027400478312256339496385275012465661575576202252063145698732079880294664220579764848767704076761853197216563262660046602703973050798218246170835962005598561669706844469447435461092542265792444947706769615695252256130901271870341005768912974433684521436211263358097522726462083917939091760026658925757076733484173202927141441492573799914240222628795405623953109131594523623353044898339481494120112723445689647986475279242446083151413667587008191682564376412347964146113898565886683139407005941383669325997475076910488086663256335689181157957571445067490187939553165903773554290260531009121879044170766615232300936675369451260747671432073394867530820527479172464106442450727640226503746586340279816318821395210726268291535648506190714616083163403189943334431056876038286530365757187367147446004855912033137386225053275419626102417236133948503",
	 "1095121115716677802856811290392395128588168592409109494900178008967955253005183831872715423151551999734857184538199864469605657805519106717529655044054833197687459782636297255219742994736751541815269727940751860670268774903340296040006114013971309257028332849679096824800250742691718610670812374272414086863715763724622797509437062518082383056050144624962776302147890521249477060215148275163688301275847155316042279405557632639366066847442861422164832655874655824221577849928863023018366835675399949740429332468186340518172487073360822220449055340582568461568645259954873303616953776393853174845132081121976327462740354930744487429617202585015510744298530101547706821590188733515880733527449780963163909830077616357506845523215289297624086914545378511082534229620116563260168494523906566709418166011112754529766183554579321224940951177394088465596712620076240067370589036924024728375076210477267488679008016579588696191194060127319035195370137160936882402244399699172017835144537488486396906144217720028992863941288217185353914991583400421682751000603596655790990815525126154394344641336397793791497068253936771017031980867706707490224041075826337383538651825493679503771934836094655802776331664261631740148281763487765852746577808019633679",

	 /* generic unrestricted moduli */
	 "17933601194860113372237070562165128350027320072176844226673287945873370751245439587792371960615073855669274087805055507977323024886880985062002853331424203",
	 "2893527720709661239493896562339544088620375736490408468011883030469939904368086092336458298221245707898933583190713188177399401852627749210994595974791782790253946539043962213027074922559572312141181787434278708783207966459019479487",
	 "347743159439876626079252796797422223177535447388206607607181663903045907591201940478223621722118173270898487582987137708656414344685816179420855160986340457973820182883508387588163122354089264395604796675278966117567294812714812796820596564876450716066283126720010859041484786529056457896367683122960411136319",
	 "47266428956356393164697365098120418976400602706072312735924071745438532218237979333351774907308168340693326687317443721193266215155735814510792148768576498491199122744351399489453533553203833318691678263241941706256996197460424029012419012634671862283532342656309677173602509498417976091509154360039893165037637034737020327399910409885798185771003505320583967737293415979917317338985837385734747478364242020380416892056650841470869294527543597349250299539682430605173321029026555546832473048600327036845781970289288898317888427517364945316709081173840186150794397479045034008257793436817683392375274635794835245695887",
	 "436463808505957768574894870394349739623346440601945961161254440072143298152040105676491048248110146278752857839930515766167441407021501229924721335644557342265864606569000117714935185566842453630868849121480179691838399545644365571106757731317371758557990781880691336695584799313313687287468894148823761785582982549586183756806449017542622267874275103877481475534991201849912222670102069951687572917937634467778042874315463238062009202992087620963771759666448266532858079402669920025224220613419441069718482837399612644978839925207109870840278194042158748845445131729137117098529028886770063736487420613144045836803985635654192482395882603511950547826439092832800532152534003936926017612446606135655146445620623395788978726744728503058670046885876251527122350275750995227",
	 "11424167473351836398078306042624362277956429440521137061889702611766348760692206243140413411077394583180726863277012016602279290144126785129569474909173584789822341986742719230331946072730319555984484911716797058875905400999504305877245849119687509023232790273637466821052576859232452982061831009770786031785669030271542286603956118755585683996118896215213488875253101894663403069677745948305893849505434201763745232895780711972432011344857521691017896316861403206449421332243658855453435784006517202894181640562433575390821384210960117518650374602256601091379644034244332285065935413233557998331562749140202965844219336298970011513882564935538704289446968322281451907487362046511461221329799897350993370560697505809686438782036235372137015731304779072430260986460269894522159103008260495503005267165927542949439526272736586626709581721032189532726389643625590680105784844246152702670169304203783072275089194754889511973916207",
	 "1214855636816562637502584060163403830270705000634713483015101384881871978446801224798536155406895823305035467591632531067547890948695117172076954220727075688048751022421198712032848890056357845974246560748347918630050853933697792254955890439720297560693579400297062396904306270145886830719309296352765295712183040773146419022875165382778007040109957609739589875590885701126197906063620133954893216612678838507540777138437797705602453719559017633986486649523611975865005712371194067612263330335590526176087004421363598470302731349138773205901447704682181517904064735636518462452242791676541725292378925568296858010151852326316777511935037531017413910506921922450666933202278489024521263798482237150056835746454842662048692127173834433089016107854491097456725016327709663199738238442164843147132789153725513257167915555162094970853584447993125488607696008169807374736711297007473812256272245489405898470297178738029484459690836250560495461579533254473316340608217876781986188705928270735695752830825527963838355419762516246028680280988020401914551825487349990306976304093109384451438813251211051597392127491464898797406789175453067960072008590614886532333015881171367104445044718144312416815712216611576221546455968770801413440778423979",
	 NULL
      };
      log = fopen("logs/expt.log", "w");
      logb = fopen("logs/expt_dr.log", "w");
      logc = fopen("logs/expt_2k.log", "w");
      logd = fopen("logs/expt_2kl.log", "w");
      for (n = 0; primes[n]; n++) {
	 SLEEP;
	 mp_read_radix(&a, primes[n], 10);
	 mp_zero(&b);
	 for (rr = 0; rr < (unsigned) mp_count_bits(&a); rr++) {
	    mp_mul_2(&b, &b);
	    b.dp[0] |= lbit();
	    b.used += 1;
	 }
	 mp_sub_d(&a, 1, &c);
	 mp_mod(&b, &c, &b);
	 mp_set(&c, 3);
	 rr = 0;
	 tt = -1;
	 do {
	    gg = TIMFUNC();
	    DO(mp_exptmod(&c, &b, &a, &d));
	    gg = (TIMFUNC() - gg) >> 1;
	    if (tt > gg)
	       tt = gg;
	 } while (++rr < 10);
	 mp_sub_d(&a, 1, &e);
	 mp_sub(&e, &b, &b);
	 mp_exptmod(&c, &b, &a, &e);	/* c^(p-1-b) mod a */
	 mp_mulmod(&e, &d, &a, &d);	/* c^b * c^(p-1-b) == c^p-1 == 1 */
	 if (mp_cmp_d(&d, 1)) {
	    printf("Different (%d)!!!\n", mp_count_bits(&a));
	    draw(&d);
	    exit(0);
	 }
	 printf("Exponentiating\t%4d-bit => %9llu/sec, %9llu cycles\n",
		mp_count_bits(&a), CLK_PER_SEC / tt, tt);
	 fprintf(n < 4 ? logd : (n < 9) ? logc : (n < 16) ? logb : log,
		 "%d %9llu\n", mp_count_bits(&a), tt);
      }
   }
   fclose(log);
   fclose(logb);
   fclose(logc);
   fclose(logd);

   log = fopen("logs/invmod.log", "w");
   for (cnt = 4; cnt <= 128; cnt += 4) {
      SLEEP;
      mp_rand(&a, cnt);
      mp_rand(&b, cnt);

      do {
	 mp_add_d(&b, 1, &b);
	 mp_gcd(&a, &b, &c);
      } while (mp_cmp_d(&c, 1) != MP_EQ);

      rr = 0;
      tt = -1;
      do {
	 gg = TIMFUNC();
	 DO(mp_invmod(&b, &a, &c));
	 gg = (TIMFUNC() - gg) >> 1;
	 if (tt > gg)
	    tt = gg;
      } while (++rr < 1000);
      mp_mulmod(&b, &c, &a, &d);
      if (mp_cmp_d(&d, 1) != MP_EQ) {
	 printf("Failed to invert\n");
	 return 0;
      }
      printf("Inverting mod\t%4d-bit => %9llu/sec, %9llu cycles\n",
	     mp_count_bits(&a), CLK_PER_SEC / tt, tt);
      fprintf(log, "%d %9llu\n", cnt * DIGIT_BIT, tt);
   }
   fclose(log);

   return 0;
}

256-bits (k = 36113) = 115792089237316195423570985008687907853269984665640564039457584007913129603823
512-bits (k = 38117) = 13407807929942597099574024998205846127479365820592393377723561443721764030073546976801874298166903427690031858186486050853753882811946569946433649006045979

/* Makes safe primes of a 2k nature */
#include <tommath.h>
#include <time.h>

int sizes[] = {256, 512, 768, 1024, 1536, 2048, 3072, 4096};

int main(void)
{
   char buf[2000];
   int x, y;
   mp_int q, p;
   FILE *out;
   clock_t t1;
   mp_digit z;
   
   mp_init_multi(&q, &p, NULL);
   
   out = fopen("2kprime.1", "w");
   for (x = 0; x < (int)(sizeof(sizes) / sizeof(sizes[0])); x++) {
   top:
       mp_2expt(&q, sizes[x]);
       mp_add_d(&q, 3, &q);
       z = -3;
       
       t1 = clock();
       for(;;) {
         mp_sub_d(&q, 4, &q);
         z += 4;

         if (z > MP_MASK) {
            printf("No primes of size %d found\n", sizes[x]);
            break;
         }
         
         if (clock() - t1 > CLOCKS_PER_SEC) { 
            printf("."); fflush(stdout);
//            sleep((clock() - t1 + CLOCKS_PER_SEC/2)/CLOCKS_PER_SEC);
            t1 = clock();
         }
         
         /* quick test on q */
         mp_prime_is_prime(&q, 1, &y);
         if (y == 0) {
            continue;
         }

         /* find (q-1)/2 */
         mp_sub_d(&q, 1, &p);
         mp_div_2(&p, &p);
         mp_prime_is_prime(&p, 3, &y);
         if (y == 0) {
            continue;
         }

         /* test on q */
         mp_prime_is_prime(&q, 3, &y);
         if (y == 0) {
            continue;
         }

         break;
       }
       
       if (y == 0) {
          ++sizes[x];
          goto top;
       }
       
       mp_toradix(&q, buf, 10);
       printf("\n\n%d-bits (k = %lu) = %s\n", sizes[x], z, buf);
       fprintf(out, "%d-bits (k = %lu) = %s\n", sizes[x], z, buf); fflush(out);
   }
   
   return 0;
}   

/* Makes safe primes of a DR nature */
#include <tommath.h>

int sizes[] = { 1+256/DIGIT_BIT, 1+512/DIGIT_BIT, 1+768/DIGIT_BIT, 1+1024/DIGIT_BIT, 1+2048/DIGIT_BIT, 1+4096/DIGIT_BIT };
int main(void)
{
   int res, x, y;
   char buf[4096];
   FILE *out;
   mp_int a, b;
   
   mp_init(&a);
   mp_init(&b);
   
   out = fopen("drprimes.txt", "w");
   for (x = 0; x < (int)(sizeof(sizes)/sizeof(sizes[0])); x++) {
   top:
       printf("Seeking a %d-bit safe prime\n", sizes[x] * DIGIT_BIT);
       mp_grow(&a, sizes[x]);
       mp_zero(&a);
       for (y = 1; y < sizes[x]; y++) {
           a.dp[y] = MP_MASK;
       }
       
       /* make a DR modulus */
       a.dp[0] = -1;
       a.used = sizes[x];
       
       /* now loop */
       res = 0;
       for (;;) { 
          a.dp[0] += 4;
          if (a.dp[0] >= MP_MASK) break;
          mp_prime_is_prime(&a, 1, &res);
          if (res == 0) continue;
          printf("."); fflush(stdout);
          mp_sub_d(&a, 1, &b);
          mp_div_2(&b, &b);
          mp_prime_is_prime(&b, 3, &res);  
          if (res == 0) continue;
          mp_prime_is_prime(&a, 3, &res);
          if (res == 1) break;
	}
        
        if (res != 1) {
           printf("Error not DR modulus\n"); sizes[x] += 1; goto top;
        } else {
           mp_toradix(&a, buf, 10);
           printf("\n\np == %s\n\n", buf);
           fprintf(out, "%d-bit prime:\np == %s\n\n", mp_count_bits(&a), buf); fflush(out);
        }           
   }
   fclose(out);
   
   mp_clear(&a);
   mp_clear(&b);
   
   return 0;
}

DR safe primes for 28-bit digits.

224-bit prime:
p == 26959946667150639794667015087019630673637144422540572481103341844143

532-bit prime:
p == 14059105607947488696282932836518693308967803494693489478439861164411992439598399594747002144074658928593502845729752797260025831423419686528151609940203368691747

784-bit prime:
p == 101745825697019260773923519755878567461315282017759829107608914364075275235254395622580447400994175578963163918967182013639660669771108475957692810857098847138903161308502419410142185759152435680068435915159402496058513611411688900243039

1036-bit prime:
p == 736335108039604595805923406147184530889923370574768772191969612422073040099331944991573923112581267542507986451953227192970402893063850485730703075899286013451337291468249027691733891486704001513279827771740183629161065194874727962517148100775228363421083691764065477590823919364012917984605619526140821798437127

1540-bit prime:
p == 38564998830736521417281865696453025806593491967131023221754800625044118265468851210705360385717536794615180260494208076605798671660719333199513807806252394423283413430106003596332513246682903994829528690198205120921557533726473585751382193953592127439965050261476810842071573684505878854588706623484573925925903505747545471088867712185004135201289273405614415899438276535626346098904241020877974002916168099951885406379295536200413493190419727789712076165162175783

2072-bit prime:
p == 542189391331696172661670440619180536749994166415993334151601745392193484590296600979602378676624808129613777993466242203025054573692562689251250471628358318743978285860720148446448885701001277560572526947619392551574490839286458454994488665744991822837769918095117129546414124448777033941223565831420390846864429504774477949153794689948747680362212954278693335653935890352619041936727463717926744868338358149568368643403037768649616778526013610493696186055899318268339432671541328195724261329606699831016666359440874843103020666106568222401047720269951530296879490444224546654729111504346660859907296364097126834834235287147

3080-bit prime:
p == 1487259134814709264092032648525971038895865645148901180585340454985524155135260217788758027400478312256339496385275012465661575576202252063145698732079880294664220579764848767704076761853197216563262660046602703973050798218246170835962005598561669706844469447435461092542265792444947706769615695252256130901271870341005768912974433684521436211263358097522726462083917939091760026658925757076733484173202927141441492573799914240222628795405623953109131594523623353044898339481494120112723445689647986475279242446083151413667587008191682564376412347964146113898565886683139407005941383669325997475076910488086663256335689181157957571445067490187939553165903773554290260531009121879044170766615232300936675369451260747671432073394867530820527479172464106442450727640226503746586340279816318821395210726268291535648506190714616083163403189943334431056876038286530365757187367147446004855912033137386225053275419626102417236133948503

4116-bit prime:
p == 1095121115716677802856811290392395128588168592409109494900178008967955253005183831872715423151551999734857184538199864469605657805519106717529655044054833197687459782636297255219742994736751541815269727940751860670268774903340296040006114013971309257028332849679096824800250742691718610670812374272414086863715763724622797509437062518082383056050144624962776302147890521249477060215148275163688301275847155316042279405557632639366066847442861422164832655874655824221577849928863023018366835675399949740429332468186340518172487073360822220449055340582568461568645259954873303616953776393853174845132081121976327462740354930744487429617202585015510744298530101547706821590188733515880733527449780963163909830077616357506845523215289297624086914545378511082534229620116563260168494523906566709418166011112754529766183554579321224940951177394088465596712620076240067370589036924024728375076210477267488679008016579588696191194060127319035195370137160936882402244399699172017835144537488486396906144217720028992863941288217185353914991583400421682751000603596655790990815525126154394344641336397793791497068253936771017031980867706707490224041075826337383538651825493679503771934836094655802776331664261631740148281763487765852746577808019633679

300-bit prime:
p == 2037035976334486086268445688409378161051468393665936250636140449354381298610415201576637819

540-bit prime:
p == 3599131035634557106248430806148785487095757694641533306480604458089470064537190296255232548883112685719936728506816716098566612844395439751206810991770626477344739

780-bit prime:
p == 6359114106063703798370219984742410466332205126109989319225557147754704702203399726411277962562135973685197744935448875852478791860694279747355800678568677946181447581781401213133886609947027230004277244697462656003655947791725966271167

CFLAGS += -Wall -W -Wshadow -O3 -fomit-frame-pointer -funroll-loops -I../

# default lib name (requires install with root)
# LIBNAME=-ltommath

# libname when you can't install the lib with install
LIBNAME=../libtommath.a

#provable primes
pprime: pprime.o
	$(CC) pprime.o $(LIBNAME) -o pprime

# portable [well requires clock()] tuning app
tune: tune.o
	$(CC) tune.o $(LIBNAME) -o tune
	
# same app but using RDTSC for higher precision [requires 80586+], coff based gcc installs [e.g. ming, cygwin, djgpp]
tune86: tune.c
	nasm -f coff timer.asm
	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o  $(LIBNAME) -o tune86
	
# for cygwin
tune86c: tune.c
	nasm -f gnuwin32 timer.asm
	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o  $(LIBNAME) -o tune86

#make tune86 for linux or any ELF format
tune86l: tune.c
	nasm -f elf -DUSE_ELF timer.asm
	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o $(LIBNAME) -o tune86l
        
# spits out mersenne primes
mersenne: mersenne.o
	$(CC) mersenne.o $(LIBNAME) -o mersenne

# fines DR safe primes for the given config
drprime: drprime.o
	$(CC) drprime.o $(LIBNAME) -o drprime
	
# fines 2k safe primes for the given config
2kprime: 2kprime.o
	$(CC) 2kprime.o $(LIBNAME) -o 2kprime

mont: mont.o
	$(CC) mont.o $(LIBNAME) -o mont

        
clean:
	rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime tune86 tune86l mont 2kprime pprime.dat \
         *.da *.dyn *.dpi *~

CC = icc

CFLAGS += -I../

# optimize for SPEED
#
# -mcpu= can be pentium, pentiumpro (covers PII through PIII) or pentium4
# -ax?   specifies make code specifically for ? but compatible with IA-32
# -x?    specifies compile solely for ? [not specifically IA-32 compatible]
#
# where ? is 
#   K - PIII
#   W - first P4 [Williamette]
#   N - P4 Northwood
#   P - P4 Prescott
#   B - Blend of P4 and PM [mobile]
#
# Default to just generic max opts
CFLAGS += -O3 -xP -ip

# default lib name (requires install with root)
# LIBNAME=-ltommath

# libname when you can't install the lib with install
LIBNAME=../libtommath.a

#provable primes
pprime: pprime.o
	$(CC) pprime.o $(LIBNAME) -o pprime

# portable [well requires clock()] tuning app
tune: tune.o
	$(CC) tune.o $(LIBNAME) -o tune
	
# same app but using RDTSC for higher precision [requires 80586+], coff based gcc installs [e.g. ming, cygwin, djgpp]
tune86: tune.c
	nasm -f coff timer.asm
	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o  $(LIBNAME) -o tune86
	
# for cygwin
tune86c: tune.c
	nasm -f gnuwin32 timer.asm
	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o  $(LIBNAME) -o tune86

#make tune86 for linux or any ELF format
tune86l: tune.c
	nasm -f elf -DUSE_ELF timer.asm
	$(CC) -DX86_TIMER $(CFLAGS) tune.c timer.o $(LIBNAME) -o tune86l
        
# spits out mersenne primes
mersenne: mersenne.o
	$(CC) mersenne.o $(LIBNAME) -o mersenne

# fines DR safe primes for the given config
drprime: drprime.o
	$(CC) drprime.o $(LIBNAME) -o drprime
	
# fines 2k safe primes for the given config
2kprime: 2kprime.o
	$(CC) 2kprime.o $(LIBNAME) -o 2kprime

mont: mont.o
	$(CC) mont.o $(LIBNAME) -o mont

        
clean:
	rm -f *.log *.o *.obj *.exe pprime tune mersenne drprime tune86 tune86l mont 2kprime pprime.dat *.il

#MSVC Makefile
#
#Tom St Denis

CFLAGS = /I../ /Ox /DWIN32 /W3

pprime: pprime.obj
	cl pprime.obj ../tommath.lib 

mersenne: mersenne.obj
	cl mersenne.obj ../tommath.lib
	
tune: tune.obj
	cl tune.obj ../tommath.lib

mont: mont.obj
	cl mont.obj ../tommath.lib
	
drprime: drprime.obj
	cl drprime.obj ../tommath.lib

2kprime: 2kprime.obj
	cl 2kprime.obj ../tommath.lib

/* Finds Mersenne primes using the Lucas-Lehmer test 
 *
 * Tom St Denis, [email protected]
 */
#include <time.h>
#include <tommath.h>

int
is_mersenne (long s, int *pp)
{
  mp_int  n, u;
  int     res, k;
  
  *pp = 0;

  if ((res = mp_init (&n)) != MP_OKAY) {
    return res;
  }

  if ((res = mp_init (&u)) != MP_OKAY) {
    goto LBL_N;
  }

  /* n = 2^s - 1 */
  if ((res = mp_2expt(&n, s)) != MP_OKAY) {
     goto LBL_MU;
  }
  if ((res = mp_sub_d (&n, 1, &n)) != MP_OKAY) {
    goto LBL_MU;
  }

  /* set u=4 */
  mp_set (&u, 4);

  /* for k=1 to s-2 do */
  for (k = 1; k <= s - 2; k++) {
    /* u = u^2 - 2 mod n */
    if ((res = mp_sqr (&u, &u)) != MP_OKAY) {
      goto LBL_MU;
    }
    if ((res = mp_sub_d (&u, 2, &u)) != MP_OKAY) {
      goto LBL_MU;
    }

    /* make sure u is positive */
    while (u.sign == MP_NEG) {
      if ((res = mp_add (&u, &n, &u)) != MP_OKAY) {
         goto LBL_MU;
      }
    }

    /* reduce */
    if ((res = mp_reduce_2k (&u, &n, 1)) != MP_OKAY) {
      goto LBL_MU;
    }
  }

  /* if u == 0 then its prime */
  if (mp_iszero (&u) == 1) {
    mp_prime_is_prime(&n, 8, pp);
  if (*pp != 1) printf("FAILURE\n");
  }

  res = MP_OKAY;
LBL_MU:mp_clear (&u);
LBL_N:mp_clear (&n);
  return res;
}

/* square root of a long < 65536 */
long
i_sqrt (long x)
{
  long    x1, x2;

  x2 = 16;
  do {
    x1 = x2;
    x2 = x1 - ((x1 * x1) - x) / (2 * x1);
  } while (x1 != x2);

  if (x1 * x1 > x) {
    --x1;
  }

  return x1;
}

/* is the long prime by brute force */
int
isprime (long k)
{
  long    y, z;

  y = i_sqrt (k);
  for (z = 2; z <= y; z++) {
    if ((k % z) == 0)
      return 0;
  }
  return 1;
}


int
main (void)
{
  int     pp;
  long    k;
  clock_t tt;

  k = 3;

  for (;;) {
    /* start time */
    tt = clock ();

    /* test if 2^k - 1 is prime */
    if (is_mersenne (k, &pp) != MP_OKAY) {
      printf ("Whoa error\n");
      return -1;
    }

    if (pp == 1) {
      /* count time */
      tt = clock () - tt;

      /* display if prime */
      printf ("2^%-5ld - 1 is prime, test took %ld ticks\n", k, tt);
    }

    /* goto next odd exponent */
    k += 2;

    /* but make sure its prime */
    while (isprime (k) == 0) {
      k += 2;
    }
  }
  return 0;
}

/* tests the montgomery routines */
#include <tommath.h>

int main(void)
{
   mp_int modulus, R, p, pp;
   mp_digit mp;
   long x, y;

   srand(time(NULL));
   mp_init_multi(&modulus, &R, &p, &pp, NULL);

   /* loop through various sizes */
   for (x = 4; x < 256; x++) {
       printf("DIGITS == %3ld...", x); fflush(stdout);
       
       /* make up the odd modulus */
       mp_rand(&modulus, x);
       modulus.dp[0] |= 1;
       
       /* now find the R value */
       mp_montgomery_calc_normalization(&R, &modulus);
       mp_montgomery_setup(&modulus, &mp);
       
       /* now run through a bunch tests */
       for (y = 0; y < 1000; y++) {
           mp_rand(&p, x/2);        /* p = random */
           mp_mul(&p, &R, &pp);     /* pp = R * p */
           mp_montgomery_reduce(&pp, &modulus, mp);
           
           /* should be equal to p */
           if (mp_cmp(&pp, &p) != MP_EQ) {
              printf("FAILURE!\n");
              exit(-1);
           }
       }
       printf("PASSED\n");
    }
    
    return 0;
}

/* Generates provable primes
 *
 * See http://gmail.com:8080/papers/pp.pdf for more info.
 *
 * Tom St Denis, [email protected], http://tom.gmail.com
 */
#include <time.h>
#include "tommath.h"

int   n_prime;
FILE *primes;

/* fast square root */
static  mp_digit
i_sqrt (mp_word x)
{
  mp_word x1, x2;

  x2 = x;
  do {
    x1 = x2;
    x2 = x1 - ((x1 * x1) - x) / (2 * x1);
  } while (x1 != x2);

  if (x1 * x1 > x) {
    --x1;
  }

  return x1;
}


/* generates a prime digit */
static void gen_prime (void)
{
  mp_digit r, x, y, next;
  FILE *out;

  out = fopen("pprime.dat", "wb");

  /* write first set of primes */
  r = 3; fwrite(&r, 1, sizeof(mp_digit), out);
  r = 5; fwrite(&r, 1, sizeof(mp_digit), out);
  r = 7; fwrite(&r, 1, sizeof(mp_digit), out);
  r = 11; fwrite(&r, 1, sizeof(mp_digit), out);
  r = 13; fwrite(&r, 1, sizeof(mp_digit), out);
  r = 17; fwrite(&r, 1, sizeof(mp_digit), out);
  r = 19; fwrite(&r, 1, sizeof(mp_digit), out);
  r = 23; fwrite(&r, 1, sizeof(mp_digit), out);
  r = 29; fwrite(&r, 1, sizeof(mp_digit), out);
  r = 31; fwrite(&r, 1, sizeof(mp_digit), out);

  /* get square root, since if 'r' is composite its factors must be < than this */
  y = i_sqrt (r);
  next = (y + 1) * (y + 1);

  for (;;) {
  do {
    r += 2;			/* next candidate */
    r &= MP_MASK;
    if (r < 31) break;

    /* update sqrt ? */
    if (next <= r) {
      ++y;
      next = (y + 1) * (y + 1);
    }

    /* loop if divisible by 3,5,7,11,13,17,19,23,29  */
    if ((r % 3) == 0) {
      x = 0;
      continue;
    }
    if ((r % 5) == 0) {
      x = 0;
      continue;
    }
    if ((r % 7) == 0) {
      x = 0;
      continue;
    }
    if ((r % 11) == 0) {
      x = 0;
      continue;
    }
    if ((r % 13) == 0) {
      x = 0;
      continue;
    }
    if ((r % 17) == 0) {
      x = 0;
      continue;
    }
    if ((r % 19) == 0) {
      x = 0;
      continue;
    }
    if ((r % 23) == 0) {
      x = 0;
      continue;
    }
    if ((r % 29) == 0) {
      x = 0;
      continue;
    }

    /* now check if r is divisible by x + k={1,7,11,13,17,19,23,29} */
    for (x = 30; x <= y; x += 30) {
      if ((r % (x + 1)) == 0) {
	x = 0;
	break;
      }
      if ((r % (x + 7)) == 0) {
	x = 0;
	break;
      }
      if ((r % (x + 11)) == 0) {
	x = 0;
	break;
      }
      if ((r % (x + 13)) == 0) {
	x = 0;
	break;
      }
      if ((r % (x + 17)) == 0) {
	x = 0;
	break;
      }
      if ((r % (x + 19)) == 0) {
	x = 0;
	break;
      }
      if ((r % (x + 23)) == 0) {
	x = 0;
	break;
      }
      if ((r % (x + 29)) == 0) {
	x = 0;
	break;
      }
    }
  } while (x == 0);
  if (r > 31) { fwrite(&r, 1, sizeof(mp_digit), out); printf("%9d\r", r); fflush(stdout); }
  if (r < 31) break;
  }

  fclose(out);
}

void load_tab(void)
{
   primes = fopen("pprime.dat", "rb");
   if (primes == NULL) {
      gen_prime();
      primes = fopen("pprime.dat", "rb");
   }
   fseek(primes, 0, SEEK_END);
   n_prime = ftell(primes) / sizeof(mp_digit);
}

mp_digit prime_digit(void)
{
   int n;
   mp_digit d;

   n = abs(rand()) % n_prime;
   fseek(primes, n * sizeof(mp_digit), SEEK_SET);
   fread(&d, 1, sizeof(mp_digit), primes);
   return d;
}


/* makes a prime of at least k bits */
int
pprime (int k, int li, mp_int * p, mp_int * q)
{
  mp_int  a, b, c, n, x, y, z, v;
  int     res, ii;
  static const mp_digit bases[] = { 2, 3, 5, 7, 11, 13, 17, 19 };

  /* single digit ? */
  if (k <= (int) DIGIT_BIT) {
    mp_set (p, prime_digit ());
    return MP_OKAY;
  }

  if ((res = mp_init (&c)) != MP_OKAY) {
    return res;
  }

  if ((res = mp_init (&v)) != MP_OKAY) {
    goto LBL_C;
  }

  /* product of first 50 primes */
  if ((res =
       mp_read_radix (&v,
		      "19078266889580195013601891820992757757219839668357012055907516904309700014933909014729740190",
		      10)) != MP_OKAY) {
    goto LBL_V;
  }

  if ((res = mp_init (&a)) != MP_OKAY) {
    goto LBL_V;
  }

  /* set the prime */
  mp_set (&a, prime_digit ());

  if ((res = mp_init (&b)) != MP_OKAY) {
    goto LBL_A;
  }

  if ((res = mp_init (&n)) != MP_OKAY) {
    goto LBL_B;
  }

  if ((res = mp_init (&x)) != MP_OKAY) {
    goto LBL_N;
  }

  if ((res = mp_init (&y)) != MP_OKAY) {
    goto LBL_X;
  }

  if ((res = mp_init (&z)) != MP_OKAY) {
    goto LBL_Y;
  }

  /* now loop making the single digit */
  while (mp_count_bits (&a) < k) {
    fprintf (stderr, "prime has %4d bits left\r", k - mp_count_bits (&a));
    fflush (stderr);
  top:
    mp_set (&b, prime_digit ());

    /* now compute z = a * b * 2 */
    if ((res = mp_mul (&a, &b, &z)) != MP_OKAY) {	/* z = a * b */
      goto LBL_Z;
    }

    if ((res = mp_copy (&z, &c)) != MP_OKAY) {	/* c = a * b */
      goto LBL_Z;
    }

    if ((res = mp_mul_2 (&z, &z)) != MP_OKAY) {	/* z = 2 * a * b */
      goto LBL_Z;
    }

    /* n = z + 1 */
    if ((res = mp_add_d (&z, 1, &n)) != MP_OKAY) {	/* n = z + 1 */
      goto LBL_Z;
    }

    /* check (n, v) == 1 */
    if ((res = mp_gcd (&n, &v, &y)) != MP_OKAY) {	/* y = (n, v) */
      goto LBL_Z;
    }

    if (mp_cmp_d (&y, 1) != MP_EQ)
      goto top;

    /* now try base x=bases[ii]  */
    for (ii = 0; ii < li; ii++) {
      mp_set (&x, bases[ii]);

      /* compute x^a mod n */
      if ((res = mp_exptmod (&x, &a, &n, &y)) != MP_OKAY) {	/* y = x^a mod n */
	goto LBL_Z;
      }

      /* if y == 1 loop */
      if (mp_cmp_d (&y, 1) == MP_EQ)
	continue;

      /* now x^2a mod n */
      if ((res = mp_sqrmod (&y, &n, &y)) != MP_OKAY) {	/* y = x^2a mod n */
	goto LBL_Z;
      }

      if (mp_cmp_d (&y, 1) == MP_EQ)
	continue;

      /* compute x^b mod n */
      if ((res = mp_exptmod (&x, &b, &n, &y)) != MP_OKAY) {	/* y = x^b mod n */
	goto LBL_Z;
      }

      /* if y == 1 loop */
      if (mp_cmp_d (&y, 1) == MP_EQ)
	continue;

      /* now x^2b mod n */
      if ((res = mp_sqrmod (&y, &n, &y)) != MP_OKAY) {	/* y = x^2b mod n */
	goto LBL_Z;
      }

      if (mp_cmp_d (&y, 1) == MP_EQ)
	continue;

      /* compute x^c mod n == x^ab mod n */
      if ((res = mp_exptmod (&x, &c, &n, &y)) != MP_OKAY) {	/* y = x^ab mod n */
	goto LBL_Z;
      }

      /* if y == 1 loop */
      if (mp_cmp_d (&y, 1) == MP_EQ)
	continue;

      /* now compute (x^c mod n)^2 */
      if ((res = mp_sqrmod (&y, &n, &y)) != MP_OKAY) {	/* y = x^2ab mod n */
	goto LBL_Z;
      }

      /* y should be 1 */
      if (mp_cmp_d (&y, 1) != MP_EQ)
	continue;
      break;
    }

    /* no bases worked? */
    if (ii == li)
      goto top;

{
   char buf[4096];

   mp_toradix(&n, buf, 10);
   printf("Certificate of primality for:\n%s\n\n", buf);
   mp_toradix(&a, buf, 10);
   printf("A == \n%s\n\n", buf);
   mp_toradix(&b, buf, 10);
   printf("B == \n%s\n\nG == %d\n", buf, bases[ii]);
   printf("----------------------------------------------------------------\n");
}

    /* a = n */
    mp_copy (&n, &a);
  }

  /* get q to be the order of the large prime subgroup */
  mp_sub_d (&n, 1, q);
  mp_div_2 (q, q);
  mp_div (q, &b, q, NULL);

  mp_exch (&n, p);

  res = MP_OKAY;
LBL_Z:mp_clear (&z);
LBL_Y:mp_clear (&y);
LBL_X:mp_clear (&x);
LBL_N:mp_clear (&n);
LBL_B:mp_clear (&b);
LBL_A:mp_clear (&a);
LBL_V:mp_clear (&v);
LBL_C:mp_clear (&c);
  return res;
}


int
main (void)
{
  mp_int  p, q;
  char    buf[4096];
  int     k, li;
  clock_t t1;

  srand (time (NULL));
  load_tab();

  printf ("Enter # of bits: \n");
  fgets (buf, sizeof (buf), stdin);
  sscanf (buf, "%d", &k);

  printf ("Enter number of bases to try (1 to 8):\n");
  fgets (buf, sizeof (buf), stdin);
  sscanf (buf, "%d", &li);


  mp_init (&p);
  mp_init (&q);

  t1 = clock ();
  pprime (k, li, &p, &q);
  t1 = clock () - t1;

  printf ("\n\nTook %ld ticks, %d bits\n", t1, mp_count_bits (&p));

  mp_toradix (&p, buf, 10);
  printf ("P == %s\n", buf);
  mp_toradix (&q, buf, 10);
  printf ("Q == %s\n", buf);

  return 0;
}

Enter # of bits: 
Enter number of bases to try (1 to 8):
Certificate of primality for:
36360080703173363

A == 
89963569

B == 
202082249

G == 2
----------------------------------------------------------------
Certificate of primality for:
4851595597739856136987139

A == 
36360080703173363

B == 
66715963

G == 2
----------------------------------------------------------------
Certificate of primality for:
19550639734462621430325731591027

A == 
4851595597739856136987139

B == 
2014867

G == 2
----------------------------------------------------------------
Certificate of primality for:
10409036141344317165691858509923818734539

A == 
19550639734462621430325731591027

B == 
266207047

G == 2
----------------------------------------------------------------
Certificate of primality for:
1049829549988285012736475602118094726647504414203

A == 
10409036141344317165691858509923818734539

B == 
50428759

G == 2
----------------------------------------------------------------
Certificate of primality for:
77194737385528288387712399596835459931920358844586615003

A == 
1049829549988285012736475602118094726647504414203

B == 
36765367

G == 2
----------------------------------------------------------------
Certificate of primality for:
35663756695365208574443215955488689578374232732893628896541201763

A == 
77194737385528288387712399596835459931920358844586615003

B == 
230998627

G == 2
----------------------------------------------------------------
Certificate of primality for:
16711831463502165169495622246023119698415848120292671294127567620396469803

A == 
35663756695365208574443215955488689578374232732893628896541201763

B == 
234297127

G == 2
----------------------------------------------------------------
Certificate of primality for:
6163534781560285962890718925972249753147470953579266394395432475622345597103528739

A == 
16711831463502165169495622246023119698415848120292671294127567620396469803

B == 
184406323

G == 2
----------------------------------------------------------------
Certificate of primality for:
814258256205243497704094951432575867360065658372158511036259934640748088306764553488803787

A == 
6163534781560285962890718925972249753147470953579266394395432475622345597103528739

B == 
66054487

G == 2
----------------------------------------------------------------
Certificate of primality for:
176469695533271657902814176811660357049007467856432383037590673407330246967781451723764079581998187

A == 
814258256205243497704094951432575867360065658372158511036259934640748088306764553488803787

B == 
108362239

G == 2
----------------------------------------------------------------
Certificate of primality for:
44924492859445516541759485198544012102424796403707253610035148063863073596051272171194806669756971406400419

A == 
176469695533271657902814176811660357049007467856432383037590673407330246967781451723764079581998187

B == 
127286707

G == 2
----------------------------------------------------------------
Certificate of primality for:
20600996927219343383225424320134474929609459588323857796871086845924186191561749519858600696159932468024710985371059

A == 
44924492859445516541759485198544012102424796403707253610035148063863073596051272171194806669756971406400419

B == 
229284691

G == 2
----------------------------------------------------------------
Certificate of primality for:
6295696427695493110141186605837397185848992307978456138112526915330347715236378041486547994708748840844217371233735072572979

A == 
20600996927219343383225424320134474929609459588323857796871086845924186191561749519858600696159932468024710985371059

B == 
152800771

G == 2
----------------------------------------------------------------
Certificate of primality for:
3104984078042317488749073016454213579257792635142218294052134804187631661145261015102617582090263808696699966840735333252107678792123

A == 
6295696427695493110141186605837397185848992307978456138112526915330347715236378041486547994708748840844217371233735072572979

B == 
246595759

G == 2
----------------------------------------------------------------
Certificate of primality for:
26405175827665701256325699315126705508919255051121452292124404943796947287968603975320562847910946802396632302209435206627913466015741799499

A == 
3104984078042317488749073016454213579257792635142218294052134804187631661145261015102617582090263808696699966840735333252107678792123

B == 
4252063

G == 2
----------------------------------------------------------------
Certificate of primality for:
11122146237908413610034600609460545703591095894418599759742741406628055069007082998134905595800236452010905900391505454890446585211975124558601770163

A == 
26405175827665701256325699315126705508919255051121452292124404943796947287968603975320562847910946802396632302209435206627913466015741799499

B == 
210605419

G == 2
----------------------------------------------------------------
Certificate of primality for:
1649861642047798890580354082088712649911849362201343649289384923147797960364736011515757482030049342943790127685185806092659832129486307035500638595572396187

A == 
11122146237908413610034600609460545703591095894418599759742741406628055069007082998134905595800236452010905900391505454890446585211975124558601770163

B == 
74170111

G == 2
----------------------------------------------------------------
Certificate of primality for:
857983367126266717607389719637086684134462613006415859877666235955788392464081914127715967940968197765042399904117392707518175220864852816390004264107201177394565363

A == 
1649861642047798890580354082088712649911849362201343649289384923147797960364736011515757482030049342943790127685185806092659832129486307035500638595572396187

B == 
260016763

G == 2
----------------------------------------------------------------
Certificate of primality for:
175995909353623703257072120479340610010337144085688850745292031336724691277374210929188442230237711063783727092685448718515661641054886101716698390145283196296702450566161283

A == 
857983367126266717607389719637086684134462613006415859877666235955788392464081914127715967940968197765042399904117392707518175220864852816390004264107201177394565363

B == 
102563707

G == 2
----------------------------------------------------------------
Certificate of primality for:
48486002551155667224487059713350447239190772068092630563272168418880661006593537218144160068395218642353495339720640699721703003648144463556291315694787862009052641640656933232794283

A == 
175995909353623703257072120479340610010337144085688850745292031336724691277374210929188442230237711063783727092685448718515661641054886101716698390145283196296702450566161283

B == 
137747527

G == 2
----------------------------------------------------------------
Certificate of primality for:
13156468011529105025061495011938518171328604045212410096476697450506055664012861932372156505805788068791146986282263016790631108386790291275939575123375304599622623328517354163964228279867403

A == 
48486002551155667224487059713350447239190772068092630563272168418880661006593537218144160068395218642353495339720640699721703003648144463556291315694787862009052641640656933232794283

B == 
135672847

G == 2
----------------------------------------------------------------
Certificate of primality for:
6355194692790533601105154341731997464407930009404822926832136060319955058388106456084549316415200519472481147942263916585428906582726749131479465958107142228236909665306781538860053107680830113869123

A == 
13156468011529105025061495011938518171328604045212410096476697450506055664012861932372156505805788068791146986282263016790631108386790291275939575123375304599622623328517354163964228279867403

B == 
241523587

G == 2
----------------------------------------------------------------
Certificate of primality for:
3157116676535430302794438027544146642863331358530722860333745617571010460905857862561870488000265751138954271040017454405707755458702044884023184574412221802502351503929935224995314581932097706874819348858083

A == 
6355194692790533601105154341731997464407930009404822926832136060319955058388106456084549316415200519472481147942263916585428906582726749131479465958107142228236909665306781538860053107680830113869123

B == 
248388667

G == 2
----------------------------------------------------------------
Certificate of primality for:
390533129219992506725320633489467713907837370444962163378727819939092929448752905310115311180032249230394348337568973177802874166228132778126338883671958897238722734394783244237133367055422297736215754829839364158067

A == 
3157116676535430302794438027544146642863331358530722860333745617571010460905857862561870488000265751138954271040017454405707755458702044884023184574412221802502351503929935224995314581932097706874819348858083

B == 
61849651

G == 2
----------------------------------------------------------------
Certificate of primality for:
48583654555070224891047847050732516652910250240135992225139515777200432486685999462997073444468380434359929499498804723793106565291183220444221080449740542884172281158126259373095216435009661050109711341419005972852770440739

A == 
390533129219992506725320633489467713907837370444962163378727819939092929448752905310115311180032249230394348337568973177802874166228132778126338883671958897238722734394783244237133367055422297736215754829839364158067

B == 
62201707

G == 2
----------------------------------------------------------------
Certificate of primality for:
25733035251905120039135866524384525138869748427727001128764704499071378939227862068500633813538831598776578372709963673670934388213622433800015759585470542686333039614931682098922935087822950084908715298627996115185849260703525317419

A == 
48583654555070224891047847050732516652910250240135992225139515777200432486685999462997073444468380434359929499498804723793106565291183220444221080449740542884172281158126259373095216435009661050109711341419005972852770440739

B == 
264832231

G == 2
----------------------------------------------------------------
Certificate of primality for:
2804594464939948901906623499531073917980499195397462605359913717827014360538186518540781517129548650937632008683280555602633122170458773895504894807182664540529077836857897972175530148107545939211339044386106111633510166695386323426241809387

A == 
25733035251905120039135866524384525138869748427727001128764704499071378939227862068500633813538831598776578372709963673670934388213622433800015759585470542686333039614931682098922935087822950084908715298627996115185849260703525317419

B == 
54494047

G == 2
----------------------------------------------------------------
Certificate of primality for:
738136612083433720096707308165797114449914259256979340471077690416567237592465306112484843530074782721390528773594351482384711900456440808251196845265132086486672447136822046628407467459921823150600138073268385534588238548865012638209515923513516547

A == 
2804594464939948901906623499531073917980499195397462605359913717827014360538186518540781517129548650937632008683280555602633122170458773895504894807182664540529077836857897972175530148107545939211339044386106111633510166695386323426241809387

B == 
131594179

G == 2
----------------------------------------------------------------
Certificate of primality for:
392847529056126766528615419937165193421166694172790666626558750047057558168124866940509180171236517681470100877687445134633784815352076138790217228749332398026714192707447855731679485746120589851992221508292976900578299504461333767437280988393026452846013683

A == 
738136612083433720096707308165797114449914259256979340471077690416567237592465306112484843530074782721390528773594351482384711900456440808251196845265132086486672447136822046628407467459921823150600138073268385534588238548865012638209515923513516547

B == 
266107603

G == 2
----------------------------------------------------------------
Certificate of primality for:
168459393231883505975876919268398655632763956627405508859662408056221544310200546265681845397346956580604208064328814319465940958080244889692368602591598503944015835190587740756859842792554282496742843600573336023639256008687581291233481455395123454655488735304365627

A == 
392847529056126766528615419937165193421166694172790666626558750047057558168124866940509180171236517681470100877687445134633784815352076138790217228749332398026714192707447855731679485746120589851992221508292976900578299504461333767437280988393026452846013683

B == 
214408111

G == 2
----------------------------------------------------------------
Certificate of primality for:
14865774288636941404884923981945833072113667565310054952177860608355263252462409554658728941191929400198053290113492910272458441655458514080123870132092365833472436407455910185221474386718838138135065780840839893113912689594815485706154461164071775481134379794909690501684643

A == 
168459393231883505975876919268398655632763956627405508859662408056221544310200546265681845397346956580604208064328814319465940958080244889692368602591598503944015835190587740756859842792554282496742843600573336023639256008687581291233481455395123454655488735304365627

B == 
44122723

G == 2
----------------------------------------------------------------
Certificate of primality for:
1213301773203241614897109856134894783021668292000023984098824423682568173639394290886185366993108292039068940333907505157813934962357206131450244004178619265868614859794316361031904412926604138893775068853175215502104744339658944443630407632290152772487455298652998368296998719996019

A == 
14865774288636941404884923981945833072113667565310054952177860608355263252462409554658728941191929400198053290113492910272458441655458514080123870132092365833472436407455910185221474386718838138135065780840839893113912689594815485706154461164071775481134379794909690501684643

B == 
40808563

G == 2
----------------------------------------------------------------
Certificate of primality for:
186935245989515158127969129347464851990429060640910951266513740972248428651109062997368144722015290092846666943896556191257222521203647606911446635194198213436423080005867489516421559330500722264446765608763224572386410155413161172707802334865729654109050873820610813855041667633843601286843

A == 
1213301773203241614897109856134894783021668292000023984098824423682568173639394290886185366993108292039068940333907505157813934962357206131450244004178619265868614859794316361031904412926604138893775068853175215502104744339658944443630407632290152772487455298652998368296998719996019

B == 
77035759

G == 2
----------------------------------------------------------------
Certificate of primality for:
83142661079751490510739960019112406284111408348732592580459037404394946037094409915127399165633756159385609671956087845517678367844901424617866988187132480585966721962585586730693443536100138246516868613250009028187662080828012497191775172228832247706080044971423654632146928165751885302331924491683

A == 
186935245989515158127969129347464851990429060640910951266513740972248428651109062997368144722015290092846666943896556191257222521203647606911446635194198213436423080005867489516421559330500722264446765608763224572386410155413161172707802334865729654109050873820610813855041667633843601286843

B == 
222383587

G == 2
----------------------------------------------------------------
Certificate of primality for:
3892354773803809855317742245039794448230625839512638747643814927766738642436392673485997449586432241626440927010641564064764336402368634186618250134234189066179771240232458249806850838490410473462391401438160528157981942499581634732706904411807195259620779379274017704050790865030808501633772117217899534443

A == 
83142661079751490510739960019112406284111408348732592580459037404394946037094409915127399165633756159385609671956087845517678367844901424617866988187132480585966721962585586730693443536100138246516868613250009028187662080828012497191775172228832247706080044971423654632146928165751885302331924491683

B == 
23407687

G == 2
----------------------------------------------------------------
Certificate of primality for:
1663606652988091811284014366560171522582683318514519379924950390627250155440313691226744227787921928894551755219495501365555370027257568506349958010457682898612082048959464465369892842603765280317696116552850664773291371490339084156052244256635115997453399761029567033971998617303988376172539172702246575225837054723

A == 
3892354773803809855317742245039794448230625839512638747643814927766738642436392673485997449586432241626440927010641564064764336402368634186618250134234189066179771240232458249806850838490410473462391401438160528157981942499581634732706904411807195259620779379274017704050790865030808501633772117217899534443

B == 
213701827

G == 2
----------------------------------------------------------------


Took 33057 ticks, 1048 bits
P == 1663606652988091811284014366560171522582683318514519379924950390627250155440313691226744227787921928894551755219495501365555370027257568506349958010457682898612082048959464465369892842603765280317696116552850664773291371490339084156052244256635115997453399761029567033971998617303988376172539172702246575225837054723
Q == 3892354773803809855317742245039794448230625839512638747643814927766738642436392673485997449586432241626440927010641564064764336402368634186618250134234189066179771240232458249806850838490410473462391401438160528157981942499581634732706904411807195259620779379274017704050790865030808501633772117217899534443

Enter # of bits: 
Enter number of bases to try (1 to 8):
Certificate of primality for:
85933926807634727

A == 
253758023

B == 
169322581

G == 5
----------------------------------------------------------------
Certificate of primality for:
23930198825086241462113799

A == 
85933926807634727

B == 
139236037

G == 11
----------------------------------------------------------------
Certificate of primality for:
6401844647261612602378676572510019

A == 
23930198825086241462113799

B == 
133760791

G == 2
----------------------------------------------------------------
Certificate of primality for:
269731366027728777712034888684015329354259

A == 
6401844647261612602378676572510019

B == 
21066691

G == 2
----------------------------------------------------------------
Certificate of primality for:
37942338209025571690075025099189467992329684223707

A == 
269731366027728777712034888684015329354259

B == 
70333567

G == 2
----------------------------------------------------------------
Certificate of primality for:
15306904714258982484473490774101705363308327436988160248323

A == 
37942338209025571690075025099189467992329684223707

B == 
201712723

G == 2
----------------------------------------------------------------
Certificate of primality for:
1616744757018513392810355191503853040357155275733333124624513530099

A == 
15306904714258982484473490774101705363308327436988160248323

B == 
52810963

G == 2
----------------------------------------------------------------
Certificate of primality for:
464222094814208047161771036072622485188658077940154689939306386289983787983

A == 
1616744757018513392810355191503853040357155275733333124624513530099

B == 
143566909

G == 5
----------------------------------------------------------------
Certificate of primality for:
187429931674053784626487560729643601208757374994177258429930699354770049369025096447

A == 
464222094814208047161771036072622485188658077940154689939306386289983787983

B == 
201875281

G == 5
----------------------------------------------------------------
Certificate of primality for:
100579220846502621074093727119851331775052664444339632682598589456666938521976625305832917563

A == 
187429931674053784626487560729643601208757374994177258429930699354770049369025096447

B == 
268311523

G == 2
----------------------------------------------------------------
Certificate of primality for:
1173616081309758475197022137833792133815753368965945885089720153370737965497134878651384030219765163

A == 
100579220846502621074093727119851331775052664444339632682598589456666938521976625305832917563

B == 
5834287

G == 2
----------------------------------------------------------------
Certificate of primality for:
191456913489905913185935197655672585713573070349044195411728114905691721186574907738081340754373032735283623

A == 
1173616081309758475197022137833792133815753368965945885089720153370737965497134878651384030219765163

B == 
81567097

G == 5
----------------------------------------------------------------
Certificate of primality for:
57856530489201750164178576399448868489243874083056587683743345599898489554401618943240901541005080049321706789987519

A == 
191456913489905913185935197655672585713573070349044195411728114905691721186574907738081340754373032735283623

B == 
151095433

G == 7
----------------------------------------------------------------
Certificate of primality for:
13790529750452576698109671710773784949185621244122040804792403407272729038377767162233653248852099545134831722512085881814803

A == 
57856530489201750164178576399448868489243874083056587683743345599898489554401618943240901541005080049321706789987519

B == 
119178679

G == 2
----------------------------------------------------------------
Certificate of primality for:
7075985989000817742677547821106534174334812111605018857703825637170140040509067704269696198231266351631132464035671858077052876058979

A == 
13790529750452576698109671710773784949185621244122040804792403407272729038377767162233653248852099545134831722512085881814803

B == 
256552363

G == 2
----------------------------------------------------------------
Certificate of primality for:
1227273006232588072907488910282307435921226646895131225407452056677899411162892829564455154080310937471747140942360789623819327234258162420463

A == 
7075985989000817742677547821106534174334812111605018857703825637170140040509067704269696198231266351631132464035671858077052876058979

B == 
86720989

G == 5
----------------------------------------------------------------
Certificate of primality for:
446764896913554613686067036908702877942872355053329937790398156069936255759889884246832779737114032666318220500106499161852193765380831330106375235763

A == 
1227273006232588072907488910282307435921226646895131225407452056677899411162892829564455154080310937471747140942360789623819327234258162420463

B == 
182015287

G == 2
----------------------------------------------------------------
Certificate of primality for:
5290203010849586596974953717018896543907195901082056939587768479377028575911127944611236020459652034082251335583308070846379514569838984811187823420951275243

A == 
446764896913554613686067036908702877942872355053329937790398156069936255759889884246832779737114032666318220500106499161852193765380831330106375235763

B == 
5920567

G == 2
----------------------------------------------------------------


Took 3454 ticks, 521 bits
P == 5290203010849586596974953717018896543907195901082056939587768479377028575911127944611236020459652034082251335583308070846379514569838984811187823420951275243
Q == 446764896913554613686067036908702877942872355053329937790398156069936255759889884246832779737114032666318220500106499161852193765380831330106375235763

; x86 timer in NASM
;
; Tom St Denis, [email protected]
[bits 32]
[section .data]
time dd 0, 0

[section .text]

%ifdef USE_ELF
[global t_start]
t_start:
%else
[global _t_start]
_t_start:
%endif
   push edx
   push eax
   rdtsc
   mov [time+0],edx
   mov [time+4],eax
   pop eax
   pop edx
   ret
   
%ifdef USE_ELF
[global t_read]
t_read:
%else
[global _t_read]
_t_read:
%endif
   rdtsc
   sub eax,[time+4]
   sbb edx,[time+0]
   ret
   

/* Tune the Karatsuba parameters
 *
 * Tom St Denis, [email protected]
 */
#include <tommath.h>
#include <time.h>

/* how many times todo each size mult.  Depends on your computer.  For slow computers
 * this can be low like 5 or 10.  For fast [re: Athlon] should be 25 - 50 or so 
 */
#define TIMES (1UL<<14UL)

/* RDTSC from Scott Duplichan */
static ulong64 TIMFUNC (void)
   {
   #if defined __GNUC__
      #if defined(__i386__) || defined(__x86_64__)
         unsigned long long a;
         __asm__ __volatile__ ("rdtsc\nmovl %%eax,%0\nmovl %%edx,4+%0\n"::"m"(a):"%eax","%edx");
         return a;
      #else /* gcc-IA64 version */
         unsigned long result;
         __asm__ __volatile__("mov %0=ar.itc" : "=r"(result) :: "memory");
         while (__builtin_expect ((int) result == -1, 0))
         __asm__ __volatile__("mov %0=ar.itc" : "=r"(result) :: "memory");
         return result;
      #endif

   // Microsoft and Intel Windows compilers
   #elif defined _M_IX86
     __asm rdtsc
   #elif defined _M_AMD64
     return __rdtsc ();
   #elif defined _M_IA64
     #if defined __INTEL_COMPILER
       #include <ia64intrin.h>
     #endif
      return __getReg (3116);
   #else
     #error need rdtsc function for this build
   #endif
   }


#ifndef X86_TIMER

/* generic ISO C timer */
ulong64 LBL_T;
void t_start(void) { LBL_T = TIMFUNC(); }
ulong64 t_read(void) { return TIMFUNC() - LBL_T; }

#else
extern void t_start(void);
extern ulong64 t_read(void);
#endif

ulong64 time_mult(int size, int s)
{
  unsigned long     x;
  mp_int  a, b, c;
  ulong64 t1;

  mp_init (&a);
  mp_init (&b);
  mp_init (&c);

  mp_rand (&a, size);
  mp_rand (&b, size);

  if (s == 1) { 
      KARATSUBA_MUL_CUTOFF = size;
  } else {
      KARATSUBA_MUL_CUTOFF = 100000;
  }

  t_start();
  for (x = 0; x < TIMES; x++) {
      mp_mul(&a,&b,&c);
  }
  t1 = t_read();
  mp_clear (&a);
  mp_clear (&b);
  mp_clear (&c);
  return t1;
}

ulong64 time_sqr(int size, int s)
{
  unsigned long     x;
  mp_int  a, b;
  ulong64 t1;

  mp_init (&a);
  mp_init (&b);

  mp_rand (&a, size);

  if (s == 1) { 
      KARATSUBA_SQR_CUTOFF = size;
  } else {
      KARATSUBA_SQR_CUTOFF = 100000;
  }

  t_start();
  for (x = 0; x < TIMES; x++) {
      mp_sqr(&a,&b);
  }
  t1 = t_read();
  mp_clear (&a);
  mp_clear (&b);
  return t1;
}

int
main (void)
{
  ulong64 t1, t2;
  int x, y;

  for (x = 8; ; x += 2) { 
     t1 = time_mult(x, 0);
     t2 = time_mult(x, 1);
     printf("%d: %9llu %9llu, %9llu\n", x, t1, t2, t2 - t1);
     if (t2 < t1) break;
  }
  y = x;

  for (x = 8; ; x += 2) { 
     t1 = time_sqr(x, 0);
     t2 = time_sqr(x, 1);
     printf("%d: %9llu %9llu, %9llu\n", x, t1, t2, t2 - t1);
     if (t2 < t1) break;
  }
  printf("KARATSUBA_MUL_CUTOFF = %d\n", y);
  printf("KARATSUBA_SQR_CUTOFF = %d\n", x);

  return 0;
}

To use the pretty graphs you have to first build/run the ltmtest from the root directory of the package.  
Todo this type 

make timing ; ltmtest

in the root.  It will run for a while [about ten minutes on most PCs] and produce a series of .log files in logs/.

After doing that run "gnuplot graphs.dem" to make the PNGs.  If you managed todo that all so far just open index.html to view
them all :-)

Have fun

Tom

480        87
960       111
1440       135
1920       159
2400       200
2880       224
3360       248
3840       272
4320       296
4800       320
5280       344
5760       368
6240       392
6720       416
7200       440
7680       464

513   1435869
769   3544970
1025   7791638
2049  46902238
2561  85334899
3073 141451412
4097 308770310

607   2109225
1279  10148314
2203  34126877
3217  82716424
4253 161569606

1024   7705271
2048  34286851
4096 165207491
521   1618631

532   1928550
784   3763908
1036   7564221
1540  16566059
2072  32283784
3080  79851565
4116 157843530

set terminal png
set size 1.75
set ylabel "Cycles per Operation"
set xlabel "Operand size (bits)"

set output "addsub.png"
plot 'add.log' smooth bezier title "Addition", 'sub.log' smooth bezier title "Subtraction"

set output "mult.png"
plot 'sqr.log' smooth bezier title "Squaring (without Karatsuba)", 'sqr_kara.log' smooth bezier title "Squaring (Karatsuba)", 'mult.log' smooth bezier title "Multiplication (without Karatsuba)", 'mult_kara.log' smooth bezier title "Multiplication (Karatsuba)"

set output "expt.png"
plot 'expt.log' smooth bezier title "Exptmod (Montgomery)", 'expt_dr.log' smooth bezier title "Exptmod (Dimminished Radix)", 'expt_2k.log' smooth bezier title "Exptmod (2k Reduction)"

set output "invmod.png"
plot 'invmod.log' smooth bezier title "Modular Inverse"

<html>
<head>
<title>LibTomMath Log Plots</title>
</head>
<body>

<h1>Addition and Subtraction</h1>
<center><img src=addsub.png></center>
<hr>

<h1>Multipliers</h1>
<center><img src=mult.png></center>
<hr>

<h1>Exptmod</h1>
<center><img src=expt.png></center>
<hr>

<h1>Modular Inverse</h1>
<center><img src=invmod.png></center>
<hr>

</body>
</html>

271       555
390       855
508      1161
631      1605
749      2117
871      2687
991      3329
1108      4084
1231      4786
1351      5624
1470      6392
1586      7364
1710      8218
1830      9255
1951     10217
2067     11461
2191     12463
2308     13677
2430     14800
2551     16232
2671     17460
2791     18899
2902     20247
3028     21902
3151     23240
3267     24927
3391     26441
3511     28277
3631     29838
3749     31751
3869     33673
3989     35431
4111     37518
4231     39426
4349     41504
4471     43567
4591     45786
4711     47876
4831     50299
4951     52427
5071     54785
5189     57241
5307     59730
5431     62194
5551     64761
5670     67322
5789     70073
5907     72663
6030     75437
6151     78242
6268     81202
6389     83948
6509     86985
6631     89903
6747     93184
6869     96044
6991     99286
7109    102395
7229    105917
7351    108940
7470    112490
7589    115702
7711    119508
7831    122632
7951    126410
8071    129808
8190    133895
8311    137146
8431    141218
8549    144732
8667    149131
8790    152462
8911    156754
9030    160479
9149    165138
9271    168601
9391    173185
9511    176988
9627    181976
9751    185539
9870    190388
9991    194335
10110    199605
10228    203298

271       560
391       870
511      1159
631      1605
750      2111
871      2737
991      3361
1111      4054
1231      4778
1351      5600
1471      6404
1591      7323
1710      8255
1831      9239
1948     10257
2070     11397
2190     12531
2308     13665
2429     14870
2550     16175
2671     17539
2787     18879
2911     20350
3031     21807
3150     23415
3270     24897
3388     26567
3511     28205
3627     30076
3751     31744
3869     33657
3991     35425
4111     37522
4229     39363
4351     41503
4470     43491
4590     45827
4711     47795
4828     50166
4951     52318
5070     54911
5191     57036
5308     58237
5431     60248
5551     62678
5671     64786
5791     67294
5908     69343
6031     71607
6151     74166
6271     76590
6391     78734
6511     81175
6631     83742
6750     86403
6868     88873
6990     91150
7110     94211
7228     96922
7351     99445
7469    102216
7589    104968
7711    108113
7827    110758
7950    113714
8071    116511
8186    119643
8310    122679
8425    125581
8551    128715
8669    131778
8788    135116
8910    138138
9031    141628
9148    144754
9268    148367
9391    151551
9511    155033
9631    158652
9751    162125
9871    165248
9988    168627
10111    172427
10231    176412

265       562
389       882
509      1207
631      1572
750      1990
859      2433
991      2894
1109      3555
1230      4228
1350      5018
1471      5805
1591      6579
1709      7415
1829      8329
1949      9225
2071     10139
2188     11239
2309     12178
2431     13212
2551     14294
2671     15551
2791     16512
2911     17718
3030     18876
3150     20259
3270     21374
3391     22650
3511     23948
3631     25493
3750     26756
3870     28225
3989     29705
4110     31409
4230     32834
4351     34327
4471     35818
4591     37636
4711     39228
4830     40868
4949     42393
5070     44541
5191     46269
5310     48162
5429     49728
5548     51985
5671     53948
5791     55885
5910     57584
6031     60082
6150     62239
6270     64309
6390     66014
6511     68766
6631     71012
6750     73172
6871     74952
6991     77909
7111     80371
7231     82666
7351     84531
7469     87698
7589     90318
7711    225384
7830    232428
7950    240009
8070    246522
8190    253662
8310    260961
8431    269253
8549    275743
8671    283769
8789    290811
8911    300034
9030    306873
9149    315085
9270    323944
9390    332390
9508    337519
9631    348986
9749    356904
9871    367013
9989    373831
10108    381033
10230    393475

271       560
388       878
511      1179
629      1625
751      1988
871      2423
989      2896
1111      3561
1231      4209
1350      5015
1470      5804
1591      6556
1709      7420
1831      8263
1951      9173
2070     10153
2191     11229
2310     12167
2431     13211
2550     14309
2671     15524
2788     16525
2910     17712
3028     18822
3148     20220
3271     21343
3391     22652
3511     23944
3630     25485
3750     26778
3868     28201
3990     29653
4111     31393
4225     32841
4350     34328
4471     35786
4590     37652
4711     39245
4830     40876
4951     42433
5068     44547
5191     46321
5311     48140
5430     49727
5550     52034
5671     53954
5791     55921
5908     57597
6031     60084
6148     62226
6270     64295
6390     66045
6511     68779
6629     71003
6751     73169
6871     74992
6991     77895
7110     80376
7231     82628
7351     84468
7470     87664
7591     90284
7711     91352
7828     93995
7950     96276
8071     98691
8190    101256
8308    103631
8431    105222
8550    108343
8671    110281
8787    112764
8911    115397
9031    117690
9151    120266
9271    122715
9391    124624
9510    127937
9630    130313
9750    132914
9871    136129
9991    138517
10108    141525
10231    144225

480        94
960       116
1440       140
1920       164
2400       205
2880       229
3360       253
3840       277
4320       299
4800       321
5280       345
5760       371
6240       395
6720       419
7200       441
7680       465

const float s_logv_2[] = {
   0.000000000, 0.000000000, 1.000000000, 0.630929754, 	/*  0  1  2  3 */
   0.500000000, 0.430676558, 0.386852807, 0.356207187, 	/*  4  5  6  7 */
   0.333333333, 0.315464877, 0.301029996, 0.289064826, 	/*  8  9 10 11 */
   0.278942946, 0.270238154, 0.262649535, 0.255958025, 	/* 12 13 14 15 */
   0.250000000, 0.244650542, 0.239812467, 0.235408913, 	/* 16 17 18 19 */
   0.231378213, 0.227670249, 0.224243824, 0.221064729, 	/* 20 21 22 23 */
   0.218104292, 0.215338279, 0.212746054, 0.210309918, 	/* 24 25 26 27 */
   0.208014598, 0.205846832, 0.203795047, 0.201849087, 	/* 28 29 30 31 */
   0.200000000, 0.198239863, 0.196561632, 0.194959022, 	/* 32 33 34 35 */
   0.193426404, 0.191958720, 0.190551412, 0.189200360, 	/* 36 37 38 39 */
   0.187901825, 0.186652411, 0.185449023, 0.184288833, 	/* 40 41 42 43 */
   0.183169251, 0.182087900, 0.181042597, 0.180031327, 	/* 44 45 46 47 */
   0.179052232, 0.178103594, 0.177183820, 0.176291434, 	/* 48 49 50 51 */
   0.175425064, 0.174583430, 0.173765343, 0.172969690, 	/* 52 53 54 55 */
   0.172195434, 0.171441601, 0.170707280, 0.169991616, 	/* 56 57 58 59 */
   0.169293808, 0.168613099, 0.167948779, 0.167300179, 	/* 60 61 62 63 */
   0.166666667
};

/* Default configuration for MPI library */

#ifndef MPI_CONFIG_H_
#define MPI_CONFIG_H_

/*
  For boolean options, 
  0 = no
  1 = yes

  Other options are documented individually.

 */

#ifndef MP_IOFUNC
#define MP_IOFUNC     0  /* include mp_print() ?                */
#endif

#ifndef MP_MODARITH
#define MP_MODARITH   1  /* include modular arithmetic ?        */
#endif

#ifndef MP_NUMTH
#define MP_NUMTH      1  /* include number theoretic functions? */
#endif

#ifndef MP_LOGTAB
#define MP_LOGTAB     1  /* use table of logs instead of log()? */
#endif

#ifndef MP_MEMSET
#define MP_MEMSET     1  /* use memset() to zero buffers?       */
#endif

#ifndef MP_MEMCPY
#define MP_MEMCPY     1  /* use memcpy() to copy buffers?       */
#endif

#ifndef MP_CRYPTO
#define MP_CRYPTO     1  /* erase memory on free?               */
#endif

#ifndef MP_ARGCHK
/*
  0 = no parameter checks
  1 = runtime checks, continue execution and return an error to caller
  2 = assertions; dump core on parameter errors
 */
#define MP_ARGCHK     2  /* how to check input arguments        */
#endif

#ifndef MP_DEBUG
#define MP_DEBUG      0  /* print diagnostic output?            */
#endif

#ifndef MP_DEFPREC
#define MP_DEFPREC    64 /* default precision, in digits        */
#endif

#ifndef MP_MACRO
#define MP_MACRO      1  /* use macros for frequent calls?      */
#endif

#ifndef MP_SQUARE
#define MP_SQUARE     1  /* use separate squaring code?         */
#endif

#ifndef MP_PTAB_SIZE
/*
  When building mpprime.c, we build in a table of small prime
  values to use for primality testing.  The more you include,
  the more space they take up.  See primes.c for the possible
  values (currently 16, 32, 64, 128, 256, and 6542)
 */
#define MP_PTAB_SIZE  128  /* how many built-in primes?         */
#endif

#ifndef MP_COMPAT_MACROS
#define MP_COMPAT_MACROS 1   /* define compatibility macros?    */
#endif

#endif /* ifndef MPI_CONFIG_H_ */


/* crc==3287762869, version==2, Sat Feb 02 06:43:53 2002 */

/* Type definitions generated by 'types.pl' */
typedef char               mp_sign;
typedef unsigned short     mp_digit;  /* 2 byte type */
typedef unsigned int       mp_word;   /* 4 byte type */
typedef unsigned int       mp_size;
typedef int                mp_err;

#define MP_DIGIT_BIT       (CHAR_BIT*sizeof(mp_digit))
#define MP_DIGIT_MAX       USHRT_MAX
#define MP_WORD_BIT        (CHAR_BIT*sizeof(mp_word))
#define MP_WORD_MAX        UINT_MAX

#define MP_DIGIT_SIZE      2
#define DIGIT_FMT          "%04X"
#define RADIX              (MP_DIGIT_MAX+1)

/*
    mpi.c

    by Michael J. Fromberger <[email protected]>
    Copyright (C) 1998 Michael J. Fromberger, All Rights Reserved

    Arbitrary precision integer arithmetic library
 */

#include "mpi.h"
#include <stdlib.h>
#include <string.h>
#include <ctype.h>

#if MP_DEBUG
#include <stdio.h>

#define DIAG(T,V) {fprintf(stderr,T);mp_print(V,stderr);fputc('\n',stderr);}
#else
#define DIAG(T,V)
#endif

/* 
   If MP_LOGTAB is not defined, use the math library to compute the
   logarithms on the fly.  Otherwise, use the static table below.
   Pick which works best for your system.
 */
#if MP_LOGTAB

/* {{{ s_logv_2[] - log table for 2 in various bases */

/*
  A table of the logs of 2 for various bases (the 0 and 1 entries of
  this table are meaningless and should not be referenced).  

  This table is used to compute output lengths for the mp_toradix()
  function.  Since a number n in radix r takes up about log_r(n)
  digits, we estimate the output size by taking the least integer
  greater than log_r(n), where:

  log_r(n) = log_2(n) * log_r(2)

  This table, therefore, is a table of log_r(2) for 2 <= r <= 36,
  which are the output bases supported.  
 */

#include "logtab.h"

/* }}} */
#define LOG_V_2(R)  s_logv_2[(R)]

#else

#include <math.h>
#define LOG_V_2(R)  (log(2.0)/log(R))

#endif

/* Default precision for newly created mp_int's      */
static unsigned int s_mp_defprec = MP_DEFPREC;

/* {{{ Digit arithmetic macros */

/*
  When adding and multiplying digits, the results can be larger than
  can be contained in an mp_digit.  Thus, an mp_word is used.  These
  macros mask off the upper and lower digits of the mp_word (the
  mp_word may be more than 2 mp_digits wide, but we only concern
  ourselves with the low-order 2 mp_digits)

  If your mp_word DOES have more than 2 mp_digits, you need to
  uncomment the first line, and comment out the second.
 */

/* #define  CARRYOUT(W)  (((W)>>DIGIT_BIT)&MP_DIGIT_MAX) */
#define  CARRYOUT(W)  ((W)>>DIGIT_BIT)
#define  ACCUM(W)     ((W)&MP_DIGIT_MAX)

/* }}} */

/* {{{ Comparison constants */

#define  MP_LT       -1
#define  MP_EQ        0
#define  MP_GT        1

/* }}} */

/* {{{ Constant strings */

/* Constant strings returned by mp_strerror() */
static const char *const mp_err_string[] = {
  "unknown result code",     /* say what?            */
  "boolean true",            /* MP_OKAY, MP_YES      */
  "boolean false",           /* MP_NO                */
  "out of memory",           /* MP_MEM               */
  "argument out of range",   /* MP_RANGE             */
  "invalid input parameter", /* MP_BADARG            */
  "result is undefined"      /* MP_UNDEF             */
};

/* Value to digit maps for radix conversion   */

/* s_dmap_1 - standard digits and letters */
static const char *s_dmap_1 = 
  "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz+/";

#if 0
/* s_dmap_2 - base64 ordering for digits  */
static const char *s_dmap_2 =
  "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
#endif

/* }}} */

/* {{{ Static function declarations */

/* 
   If MP_MACRO is false, these will be defined as actual functions;
   otherwise, suitable macro definitions will be used.  This works
   around the fact that ANSI C89 doesn't support an 'inline' keyword
   (although I hear C9x will ... about bloody time).  At present, the
   macro definitions are identical to the function bodies, but they'll
   expand in place, instead of generating a function call.

   I chose these particular functions to be made into macros because
   some profiling showed they are called a lot on a typical workload,
   and yet they are primarily housekeeping.
 */
#if MP_MACRO == 0
 void     s_mp_setz(mp_digit *dp, mp_size count); /* zero digits           */
 void     s_mp_copy(mp_digit *sp, mp_digit *dp, mp_size count); /* copy    */
 void    *s_mp_alloc(size_t nb, size_t ni);       /* general allocator     */
 void     s_mp_free(void *ptr);                   /* general free function */
#else

 /* Even if these are defined as macros, we need to respect the settings
    of the MP_MEMSET and MP_MEMCPY configuration options...
  */
 #if MP_MEMSET == 0
  #define  s_mp_setz(dp, count) \
       {int ix;for(ix=0;ix<(count);ix++)(dp)[ix]=0;}
 #else
  #define  s_mp_setz(dp, count) memset(dp, 0, (count) * sizeof(mp_digit))
 #endif /* MP_MEMSET */

 #if MP_MEMCPY == 0
  #define  s_mp_copy(sp, dp, count) \
       {int ix;for(ix=0;ix<(count);ix++)(dp)[ix]=(sp)[ix];}
 #else
  #define  s_mp_copy(sp, dp, count) memcpy(dp, sp, (count) * sizeof(mp_digit))
 #endif /* MP_MEMCPY */

 #define  s_mp_alloc(nb, ni)  calloc(nb, ni)
 #define  s_mp_free(ptr) {if(ptr) free(ptr);}
#endif /* MP_MACRO */

mp_err   s_mp_grow(mp_int *mp, mp_size min);   /* increase allocated size */
mp_err   s_mp_pad(mp_int *mp, mp_size min);    /* left pad with zeroes    */

void     s_mp_clamp(mp_int *mp);               /* clip leading zeroes     */

void     s_mp_exch(mp_int *a, mp_int *b);      /* swap a and b in place   */

mp_err   s_mp_lshd(mp_int *mp, mp_size p);     /* left-shift by p digits  */
void     s_mp_rshd(mp_int *mp, mp_size p);     /* right-shift by p digits */
void     s_mp_div_2d(mp_int *mp, mp_digit d);  /* divide by 2^d in place  */
void     s_mp_mod_2d(mp_int *mp, mp_digit d);  /* modulo 2^d in place     */
mp_err   s_mp_mul_2d(mp_int *mp, mp_digit d);  /* multiply by 2^d in place*/
void     s_mp_div_2(mp_int *mp);               /* divide by 2 in place    */
mp_err   s_mp_mul_2(mp_int *mp);               /* multiply by 2 in place  */
mp_digit s_mp_norm(mp_int *a, mp_int *b);      /* normalize for division  */
mp_err   s_mp_add_d(mp_int *mp, mp_digit d);   /* unsigned digit addition */
mp_err   s_mp_sub_d(mp_int *mp, mp_digit d);   /* unsigned digit subtract */
mp_err   s_mp_mul_d(mp_int *mp, mp_digit d);   /* unsigned digit multiply */
mp_err   s_mp_div_d(mp_int *mp, mp_digit d, mp_digit *r);
		                               /* unsigned digit divide   */
mp_err   s_mp_reduce(mp_int *x, mp_int *m, mp_int *mu);
                                               /* Barrett reduction       */
mp_err   s_mp_add(mp_int *a, mp_int *b);       /* magnitude addition      */
mp_err   s_mp_sub(mp_int *a, mp_int *b);       /* magnitude subtract      */
mp_err   s_mp_mul(mp_int *a, mp_int *b);       /* magnitude multiply      */
#if 0
void     s_mp_kmul(mp_digit *a, mp_digit *b, mp_digit *out, mp_size len);
                                               /* multiply buffers in place */
#endif
#if MP_SQUARE
mp_err   s_mp_sqr(mp_int *a);                  /* magnitude square        */
#else
#define  s_mp_sqr(a) s_mp_mul(a, a)
#endif
mp_err   s_mp_div(mp_int *a, mp_int *b);       /* magnitude divide        */
mp_err   s_mp_2expt(mp_int *a, mp_digit k);    /* a = 2^k                 */
int      s_mp_cmp(mp_int *a, mp_int *b);       /* magnitude comparison    */
int      s_mp_cmp_d(mp_int *a, mp_digit d);    /* magnitude digit compare */
int      s_mp_ispow2(mp_int *v);               /* is v a power of 2?      */
int      s_mp_ispow2d(mp_digit d);             /* is d a power of 2?      */

int      s_mp_tovalue(char ch, int r);          /* convert ch to value    */
char     s_mp_todigit(int val, int r, int low); /* convert val to digit   */
int      s_mp_outlen(int bits, int r);          /* output length in bytes */

/* }}} */

/* {{{ Default precision manipulation */

unsigned int mp_get_prec(void)
{
  return s_mp_defprec;

} /* end mp_get_prec() */

void         mp_set_prec(unsigned int prec)
{
  if(prec == 0)
    s_mp_defprec = MP_DEFPREC;
  else
    s_mp_defprec = prec;

} /* end mp_set_prec() */

/* }}} */

/*------------------------------------------------------------------------*/
/* {{{ mp_init(mp) */

/*
  mp_init(mp)

  Initialize a new zero-valued mp_int.  Returns MP_OKAY if successful,
  MP_MEM if memory could not be allocated for the structure.
 */

mp_err mp_init(mp_int *mp)
{
  return mp_init_size(mp, s_mp_defprec);

} /* end mp_init() */

/* }}} */

/* {{{ mp_init_array(mp[], count) */

mp_err mp_init_array(mp_int mp[], int count)
{
  mp_err  res;
  int     pos;

  ARGCHK(mp !=NULL && count > 0, MP_BADARG);

  for(pos = 0; pos < count; ++pos) {
    if((res = mp_init(&mp[pos])) != MP_OKAY)
      goto CLEANUP;
  }

  return MP_OKAY;

 CLEANUP:
  while(--pos >= 0) 
    mp_clear(&mp[pos]);

  return res;

} /* end mp_init_array() */

/* }}} */

/* {{{ mp_init_size(mp, prec) */

/*
  mp_init_size(mp, prec)

  Initialize a new zero-valued mp_int with at least the given
  precision; returns MP_OKAY if successful, or MP_MEM if memory could
  not be allocated for the structure.
 */

mp_err mp_init_size(mp_int *mp, mp_size prec)
{
  ARGCHK(mp != NULL && prec > 0, MP_BADARG);

  if((DIGITS(mp) = s_mp_alloc(prec, sizeof(mp_digit))) == NULL)
    return MP_MEM;

  SIGN(mp) = MP_ZPOS;
  USED(mp) = 1;
  ALLOC(mp) = prec;

  return MP_OKAY;

} /* end mp_init_size() */

/* }}} */

/* {{{ mp_init_copy(mp, from) */

/*
  mp_init_copy(mp, from)

  Initialize mp as an exact copy of from.  Returns MP_OKAY if
  successful, MP_MEM if memory could not be allocated for the new
  structure.
 */

mp_err mp_init_copy(mp_int *mp, mp_int *from)
{
  ARGCHK(mp != NULL && from != NULL, MP_BADARG);

  if(mp == from)
    return MP_OKAY;

  if((DIGITS(mp) = s_mp_alloc(USED(from), sizeof(mp_digit))) == NULL)
    return MP_MEM;

  s_mp_copy(DIGITS(from), DIGITS(mp), USED(from));
  USED(mp) = USED(from);
  ALLOC(mp) = USED(from);
  SIGN(mp) = SIGN(from);

  return MP_OKAY;

} /* end mp_init_copy() */

/* }}} */

/* {{{ mp_copy(from, to) */

/*
  mp_copy(from, to)

  Copies the mp_int 'from' to the mp_int 'to'.  It is presumed that
  'to' has already been initialized (if not, use mp_init_copy()
  instead). If 'from' and 'to' are identical, nothing happens.
 */

mp_err mp_copy(mp_int *from, mp_int *to)
{
  ARGCHK(from != NULL && to != NULL, MP_BADARG);

  if(from == to)
    return MP_OKAY;

  { /* copy */
    mp_digit   *tmp;

    /*
      If the allocated buffer in 'to' already has enough space to hold
      all the used digits of 'from', we'll re-use it to avoid hitting
      the memory allocater more than necessary; otherwise, we'd have
      to grow anyway, so we just allocate a hunk and make the copy as
      usual
     */
    if(ALLOC(to) >= USED(from)) {
      s_mp_setz(DIGITS(to) + USED(from), ALLOC(to) - USED(from));
      s_mp_copy(DIGITS(from), DIGITS(to), USED(from));
      
    } else {
      if((tmp = s_mp_alloc(USED(from), sizeof(mp_digit))) == NULL)
	return MP_MEM;

      s_mp_copy(DIGITS(from), tmp, USED(from));

      if(DIGITS(to) != NULL) {
#if MP_CRYPTO
	s_mp_setz(DIGITS(to), ALLOC(to));
#endif
	s_mp_free(DIGITS(to));
      }

      DIGITS(to) = tmp;
      ALLOC(to) = USED(from);
    }

    /* Copy the precision and sign from the original */
    USED(to) = USED(from);
    SIGN(to) = SIGN(from);
  } /* end copy */

  return MP_OKAY;

} /* end mp_copy() */

/* }}} */

/* {{{ mp_exch(mp1, mp2) */

/*
  mp_exch(mp1, mp2)

  Exchange mp1 and mp2 without allocating any intermediate memory
  (well, unless you count the stack space needed for this call and the
  locals it creates...).  This cannot fail.
 */

void mp_exch(mp_int *mp1, mp_int *mp2)
{
#if MP_ARGCHK == 2
  assert(mp1 != NULL && mp2 != NULL);
#else
  if(mp1 == NULL || mp2 == NULL)
    return;
#endif

  s_mp_exch(mp1, mp2);

} /* end mp_exch() */

/* }}} */

/* {{{ mp_clear(mp) */

/*
  mp_clear(mp)

  Release the storage used by an mp_int, and void its fields so that
  if someone calls mp_clear() again for the same int later, we won't
  get tollchocked.
 */

void   mp_clear(mp_int *mp)
{
  if(mp == NULL)
    return;

  if(DIGITS(mp) != NULL) {
#if MP_CRYPTO
    s_mp_setz(DIGITS(mp), ALLOC(mp));
#endif
    s_mp_free(DIGITS(mp));
    DIGITS(mp) = NULL;
  }

  USED(mp) = 0;
  ALLOC(mp) = 0;

} /* end mp_clear() */

/* }}} */

/* {{{ mp_clear_array(mp[], count) */

void   mp_clear_array(mp_int mp[], int count)
{
  ARGCHK(mp != NULL && count > 0, MP_BADARG);

  while(--count >= 0) 
    mp_clear(&mp[count]);

} /* end mp_clear_array() */

/* }}} */

/* {{{ mp_zero(mp) */

/*
  mp_zero(mp) 

  Set mp to zero.  Does not change the allocated size of the structure,
  and therefore cannot fail (except on a bad argument, which we ignore)
 */
void   mp_zero(mp_int *mp)
{
  if(mp == NULL)
    return;

  s_mp_setz(DIGITS(mp), ALLOC(mp));
  USED(mp) = 1;
  SIGN(mp) = MP_ZPOS;

} /* end mp_zero() */

/* }}} */

/* {{{ mp_set(mp, d) */

void   mp_set(mp_int *mp, mp_digit d)
{
  if(mp == NULL)
    return;

  mp_zero(mp);
  DIGIT(mp, 0) = d;

} /* end mp_set() */

/* }}} */

/* {{{ mp_set_int(mp, z) */

mp_err mp_set_int(mp_int *mp, long z)
{
  int            ix;
  unsigned long  v = abs(z);
  mp_err         res;

  ARGCHK(mp != NULL, MP_BADARG);

  mp_zero(mp);
  if(z == 0)
    return MP_OKAY;  /* shortcut for zero */

  for(ix = sizeof(long) - 1; ix >= 0; ix--) {

    if((res = s_mp_mul_2d(mp, CHAR_BIT)) != MP_OKAY)
      return res;

    res = s_mp_add_d(mp, 
		     (mp_digit)((v >> (ix * CHAR_BIT)) & UCHAR_MAX));
    if(res != MP_OKAY)
      return res;

  }

  if(z < 0)
    SIGN(mp) = MP_NEG;

  return MP_OKAY;

} /* end mp_set_int() */

/* }}} */

/*------------------------------------------------------------------------*/
/* {{{ Digit arithmetic */

/* {{{ mp_add_d(a, d, b) */

/*
  mp_add_d(a, d, b)

  Compute the sum b = a + d, for a single digit d.  Respects the sign of
  its primary addend (single digits are unsigned anyway).
 */

mp_err mp_add_d(mp_int *a, mp_digit d, mp_int *b)
{
  mp_err   res = MP_OKAY;

  ARGCHK(a != NULL && b != NULL, MP_BADARG);

  if((res = mp_copy(a, b)) != MP_OKAY)
    return res;

  if(SIGN(b) == MP_ZPOS) {
    res = s_mp_add_d(b, d);
  } else if(s_mp_cmp_d(b, d) >= 0) {
    res = s_mp_sub_d(b, d);
  } else {
    SIGN(b) = MP_ZPOS;

    DIGIT(b, 0) = d - DIGIT(b, 0);
  }

  return res;

} /* end mp_add_d() */

/* }}} */

/* {{{ mp_sub_d(a, d, b) */

/*
  mp_sub_d(a, d, b)

  Compute the difference b = a - d, for a single digit d.  Respects the
  sign of its subtrahend (single digits are unsigned anyway).
 */

mp_err mp_sub_d(mp_int *a, mp_digit d, mp_int *b)
{
  mp_err   res;

  ARGCHK(a != NULL && b != NULL, MP_BADARG);

  if((res = mp_copy(a, b)) != MP_OKAY)
    return res;

  if(SIGN(b) == MP_NEG) {
    if((res = s_mp_add_d(b, d)) != MP_OKAY)
      return res;

  } else if(s_mp_cmp_d(b, d) >= 0) {
    if((res = s_mp_sub_d(b, d)) != MP_OKAY)
      return res;

  } else {
    mp_neg(b, b);

    DIGIT(b, 0) = d - DIGIT(b, 0);
    SIGN(b) = MP_NEG;
  }

  if(s_mp_cmp_d(b, 0) == 0)
    SIGN(b) = MP_ZPOS;

  return MP_OKAY;

} /* end mp_sub_d() */

/* }}} */

/* {{{ mp_mul_d(a, d, b) */

/*
  mp_mul_d(a, d, b)

  Compute the product b = a * d, for a single digit d.  Respects the sign
  of its multiplicand (single digits are unsigned anyway)
 */

mp_err mp_mul_d(mp_int *a, mp_digit d, mp_int *b)
{
  mp_err  res;

  ARGCHK(a != NULL && b != NULL, MP_BADARG);

  if(d == 0) {
    mp_zero(b);
    return MP_OKAY;
  }

  if((res = mp_copy(a, b)) != MP_OKAY)
    return res;

  res = s_mp_mul_d(b, d);

  return res;

} /* end mp_mul_d() */

/* }}} */

/* {{{ mp_mul_2(a, c) */

mp_err mp_mul_2(mp_int *a, mp_int *c)
{
  mp_err  res;

  ARGCHK(a != NULL && c != NULL, MP_BADARG);

  if((res = mp_copy(a, c)) != MP_OKAY)
    return res;

  return s_mp_mul_2(c);

} /* end mp_mul_2() */

/* }}} */

/* {{{ mp_div_d(a, d, q, r) */

/*
  mp_div_d(a, d, q, r)

  Compute the quotient q = a / d and remainder r = a mod d, for a
  single digit d.  Respects the sign of its divisor (single digits are
  unsigned anyway).
 */

mp_err mp_div_d(mp_int *a, mp_digit d, mp_int *q, mp_digit *r)
{
  mp_err   res;
  mp_digit rem;
  int      pow;

  ARGCHK(a != NULL, MP_BADARG);

  if(d == 0)
    return MP_RANGE;

  /* Shortcut for powers of two ... */
  if((pow = s_mp_ispow2d(d)) >= 0) {
    mp_digit  mask;

    mask = (1 << pow) - 1;
    rem = DIGIT(a, 0) & mask;

    if(q) {
      mp_copy(a, q);
      s_mp_div_2d(q, pow);
    }

    if(r)
      *r = rem;

    return MP_OKAY;
  }

  /*
    If the quotient is actually going to be returned, we'll try to
    avoid hitting the memory allocator by copying the dividend into it
    and doing the division there.  This can't be any _worse_ than
    always copying, and will sometimes be better (since it won't make
    another copy)

    If it's not going to be returned, we need to allocate a temporary
    to hold the quotient, which will just be discarded.
   */
  if(q) {
    if((res = mp_copy(a, q)) != MP_OKAY)
      return res;

    res = s_mp_div_d(q, d, &rem);
    if(s_mp_cmp_d(q, 0) == MP_EQ)
      SIGN(q) = MP_ZPOS;

  } else {
    mp_int  qp;

    if((res = mp_init_copy(&qp, a)) != MP_OKAY)
      return res;

    res = s_mp_div_d(&qp, d, &rem);
    if(s_mp_cmp_d(&qp, 0) == 0)
      SIGN(&qp) = MP_ZPOS;

    mp_clear(&qp);
  }

  if(r)
    *r = rem;

  return res;

} /* end mp_div_d() */

/* }}} */

/* {{{ mp_div_2(a, c) */

/*
  mp_div_2(a, c)

  Compute c = a / 2, disregarding the remainder.
 */

mp_err mp_div_2(mp_int *a, mp_int *c)
{
  mp_err  res;

  ARGCHK(a != NULL && c != NULL, MP_BADARG);

  if((res = mp_copy(a, c)) != MP_OKAY)
    return res;

  s_mp_div_2(c);

  return MP_OKAY;

} /* end mp_div_2() */

/* }}} */

/* {{{ mp_expt_d(a, d, b) */

mp_err mp_expt_d(mp_int *a, mp_digit d, mp_int *c)
{
  mp_int   s, x;
  mp_err   res;

  ARGCHK(a != NULL && c != NULL, MP_BADARG);

  if((res = mp_init(&s)) != MP_OKAY)
    return res;
  if((res = mp_init_copy(&x, a)) != MP_OKAY)
    goto X;

  DIGIT(&s, 0) = 1;

  while(d != 0) {
    if(d & 1) {
      if((res = s_mp_mul(&s, &x)) != MP_OKAY)
	goto CLEANUP;
    }

    d >>= 1;

    if((res = s_mp_sqr(&x)) != MP_OKAY)
      goto CLEANUP;
  }

  s_mp_exch(&s, c);

CLEANUP:
  mp_clear(&x);
X:
  mp_clear(&s);

  return res;

} /* end mp_expt_d() */

/* }}} */

/* }}} */

/*------------------------------------------------------------------------*/
/* {{{ Full arithmetic */

/* {{{ mp_abs(a, b) */

/*
  mp_abs(a, b)

  Compute b = |a|.  'a' and 'b' may be identical.
 */

mp_err mp_abs(mp_int *a, mp_int *b)
{
  mp_err   res;

  ARGCHK(a != NULL && b != NULL, MP_BADARG);

  if((res = mp_copy(a, b)) != MP_OKAY)
    return res;

  SIGN(b) = MP_ZPOS;

  return MP_OKAY;

} /* end mp_abs() */

/* }}} */

/* {{{ mp_neg(a, b) */

/*
  mp_neg(a, b)

  Compute b = -a.  'a' and 'b' may be identical.
 */

mp_err mp_neg(mp_int *a, mp_int *b)
{
  mp_err   res;

  ARGCHK(a != NULL && b != NULL, MP_BADARG);

  if((res = mp_copy(a, b)) != MP_OKAY)
    return res;

  if(s_mp_cmp_d(b, 0) == MP_EQ) 
    SIGN(b) = MP_ZPOS;
  else 
    SIGN(b) = (SIGN(b) == MP_NEG) ? MP_ZPOS : MP_NEG;

  return MP_OKAY;

} /* end mp_neg() */

/* }}} */

/* {{{ mp_add(a, b, c) */

/*
  mp_add(a, b, c)

  Compute c = a + b.  All parameters may be identical.
 */

mp_err mp_add(mp_int *a, mp_int *b, mp_int *c)
{
  mp_err  res;
  int     cmp;

  ARGCHK(a != NULL && b != NULL && c != NULL, MP_BADARG);

  if(SIGN(a) == SIGN(b)) { /* same sign:  add values, keep sign */

    /* Commutativity of addition lets us do this in either order,
       so we avoid having to use a temporary even if the result 
       is supposed to replace the output
     */
    if(c == b) {
      if((res = s_mp_add(c, a)) != MP_OKAY)
	return res;
    } else {
      if(c != a && (res = mp_copy(a, c)) != MP_OKAY)
	return res;

      if((res = s_mp_add(c, b)) != MP_OKAY) 
	return res;
    }

  } else if((cmp = s_mp_cmp(a, b)) > 0) {  /* different sign: a > b   */

    /* If the output is going to be clobbered, we will use a temporary
       variable; otherwise, we'll do it without touching the memory 
       allocator at all, if possible
     */
    if(c == b) {
      mp_int  tmp;

      if((res = mp_init_copy(&tmp, a)) != MP_OKAY)
	return res;
      if((res = s_mp_sub(&tmp, b)) != MP_OKAY) {
	mp_clear(&tmp);
	return res;
      }

      s_mp_exch(&tmp, c);
      mp_clear(&tmp);

    } else {

      if(c != a && (res = mp_copy(a, c)) != MP_OKAY)
	return res;
      if((res = s_mp_sub(c, b)) != MP_OKAY)
	return res;

    }

  } else if(cmp == 0) {             /* different sign, a == b   */

    mp_zero(c);
    return MP_OKAY;

  } else {                          /* different sign: a < b    */

    /* See above... */
    if(c == a) {
      mp_int  tmp;

      if((res = mp_init_copy(&tmp, b)) != MP_OKAY)
	return res;
      if((res = s_mp_sub(&tmp, a)) != MP_OKAY) {
	mp_clear(&tmp);
	return res;
      }

      s_mp_exch(&tmp, c);
      mp_clear(&tmp);

    } else {

      if(c != b && (res = mp_copy(b, c)) != MP_OKAY)
	return res;
      if((res = s_mp_sub(c, a)) != MP_OKAY)
	return res;

    }
  }

  if(USED(c) == 1 && DIGIT(c, 0) == 0)
    SIGN(c) = MP_ZPOS;

  return MP_OKAY;

} /* end mp_add() */

/* }}} */

/* {{{ mp_sub(a, b, c) */

/*
  mp_sub(a, b, c)

  Compute c = a - b.  All parameters may be identical.
 */

mp_err mp_sub(mp_int *a, mp_int *b, mp_int *c)
{
  mp_err  res;
  int     cmp;

  ARGCHK(a != NULL && b != NULL && c != NULL, MP_BADARG);

  if(SIGN(a) != SIGN(b)) {
    if(c == a) {
      if((res = s_mp_add(c, b)) != MP_OKAY)
	return res;
    } else {
      if(c != b && ((res = mp_copy(b, c)) != MP_OKAY))
	return res;
      if((res = s_mp_add(c, a)) != MP_OKAY)
	return res;
      SIGN(c) = SIGN(a);
    }

  } else if((cmp = s_mp_cmp(a, b)) > 0) { /* Same sign, a > b */
    if(c == b) {
      mp_int  tmp;

      if((res = mp_init_copy(&tmp, a)) != MP_OKAY)
	return res;
      if((res = s_mp_sub(&tmp, b)) != MP_OKAY) {
	mp_clear(&tmp);
	return res;
      }
      s_mp_exch(&tmp, c);
      mp_clear(&tmp);

    } else {
      if(c != a && ((res = mp_copy(a, c)) != MP_OKAY))
	return res;

      if((res = s_mp_sub(c, b)) != MP_OKAY)
	return res;
    }

  } else if(cmp == 0) {  /* Same sign, equal magnitude */
    mp_zero(c);
    return MP_OKAY;

  } else {               /* Same sign, b > a */
    if(c == a) {
      mp_int  tmp;

      if((res = mp_init_copy(&tmp, b)) != MP_OKAY)
	return res;

      if((res = s_mp_sub(&tmp, a)) != MP_OKAY) {
	mp_clear(&tmp);
	return res;
      }
      s_mp_exch(&tmp, c);
      mp_clear(&tmp);

    } else {
      if(c != b && ((res = mp_copy(b, c)) != MP_OKAY)) 
	return res;

      if((res = s_mp_sub(c, a)) != MP_OKAY)
	return res;
    }

    SIGN(c) = !SIGN(b);
  }

  if(USED(c) == 1 && DIGIT(c, 0) == 0)
    SIGN(c) = MP_ZPOS;

  return MP_OKAY;

} /* end mp_sub() */

/* }}} */

/* {{{ mp_mul(a, b, c) */

/*
  mp_mul(a, b, c)

  Compute c = a * b.  All parameters may be identical.
 */

mp_err mp_mul(mp_int *a, mp_int *b, mp_int *c)
{
  mp_err   res;
  mp_sign  sgn;

  ARGCHK(a != NULL && b != NULL && c != NULL, MP_BADARG);

  sgn = (SIGN(a) == SIGN(b)) ? MP_ZPOS : MP_NEG;

  if(c == b) {
    if((res = s_mp_mul(c, a)) != MP_OKAY)
      return res;

  } else {
    if((res = mp_copy(a, c)) != MP_OKAY)
      return res;

    if((res = s_mp_mul(c, b)) != MP_OKAY)
      return res;
  }
  
  if(sgn == MP_ZPOS || s_mp_cmp_d(c, 0) == MP_EQ)
    SIGN(c) = MP_ZPOS;
  else
    SIGN(c) = sgn;
  
  return MP_OKAY;

} /* end mp_mul() */

/* }}} */

/* {{{ mp_mul_2d(a, d, c) */

/*
  mp_mul_2d(a, d, c)

  Compute c = a * 2^d.  a may be the same as c.
 */

mp_err mp_mul_2d(mp_int *a, mp_digit d, mp_int *c)
{
  mp_err   res;

  ARGCHK(a != NULL && c != NULL, MP_BADARG);

  if((res = mp_copy(a, c)) != MP_OKAY)
    return res;

  if(d == 0)
    return MP_OKAY;

  return s_mp_mul_2d(c, d);

} /* end mp_mul() */

/* }}} */

/* {{{ mp_sqr(a, b) */

#if MP_SQUARE
mp_err mp_sqr(mp_int *a, mp_int *b)
{
  mp_err   res;

  ARGCHK(a != NULL && b != NULL, MP_BADARG);

  if((res = mp_copy(a, b)) != MP_OKAY)
    return res;

  if((res = s_mp_sqr(b)) != MP_OKAY)
    return res;

  SIGN(b) = MP_ZPOS;

  return MP_OKAY;

} /* end mp_sqr() */
#endif

/* }}} */

/* {{{ mp_div(a, b, q, r) */

/*
  mp_div(a, b, q, r)

  Compute q = a / b and r = a mod b.  Input parameters may be re-used
  as output parameters.  If q or r is NULL, that portion of the
  computation will be discarded (although it will still be computed)

  Pay no attention to the hacker behind the curtain.
 */

mp_err mp_div(mp_int *a, mp_int *b, mp_int *q, mp_int *r)
{
  mp_err   res;
  mp_int   qtmp, rtmp;
  int      cmp;

  ARGCHK(a != NULL && b != NULL, MP_BADARG);

  if(mp_cmp_z(b) == MP_EQ)
    return MP_RANGE;

  /* If a <= b, we can compute the solution without division, and
     avoid any memory allocation
   */
  if((cmp = s_mp_cmp(a, b)) < 0) {
    if(r) {
      if((res = mp_copy(a, r)) != MP_OKAY)
	return res;
    }

    if(q) 
      mp_zero(q);

    return MP_OKAY;

  } else if(cmp == 0) {

    /* Set quotient to 1, with appropriate sign */
    if(q) {
      int qneg = (SIGN(a) != SIGN(b));

      mp_set(q, 1);
      if(qneg)
	SIGN(q) = MP_NEG;
    }

    if(r)
      mp_zero(r);

    return MP_OKAY;
  }

  /* If we get here, it means we actually have to do some division */

  /* Set up some temporaries... */
  if((res = mp_init_copy(&qtmp, a)) != MP_OKAY)
    return res;
  if((res = mp_init_copy(&rtmp, b)) != MP_OKAY)
    goto CLEANUP;

  if((res = s_mp_div(&qtmp, &rtmp)) != MP_OKAY)
    goto CLEANUP;

  /* Compute the signs for the output  */
  SIGN(&rtmp) = SIGN(a); /* Sr = Sa              */
  if(SIGN(a) == SIGN(b))
    SIGN(&qtmp) = MP_ZPOS;  /* Sq = MP_ZPOS if Sa = Sb */
  else
    SIGN(&qtmp) = MP_NEG;   /* Sq = MP_NEG if Sa != Sb */

  if(s_mp_cmp_d(&qtmp, 0) == MP_EQ)
    SIGN(&qtmp) = MP_ZPOS;
  if(s_mp_cmp_d(&rtmp, 0) == MP_EQ)
    SIGN(&rtmp) = MP_ZPOS;

  /* Copy output, if it is needed      */
  if(q) 
    s_mp_exch(&qtmp, q);

  if(r) 
    s_mp_exch(&rtmp, r);

CLEANUP:
  mp_clear(&rtmp);
  mp_clear(&qtmp);

  return res;

} /* end mp_div() */

/* }}} */

/* {{{ mp_div_2d(a, d, q, r) */

mp_err mp_div_2d(mp_int *a, mp_digit d, mp_int *q, mp_int *r)
{
  mp_err  res;

  ARGCHK(a != NULL, MP_BADARG);

  if(q) {
    if((res = mp_copy(a, q)) != MP_OKAY)
      return res;

    s_mp_div_2d(q, d);
  }

  if(r) {
    if((res = mp_copy(a, r)) != MP_OKAY)
      return res;

    s_mp_mod_2d(r, d);
  }

  return MP_OKAY;

} /* end mp_div_2d() */

/* }}} */

/* {{{ mp_expt(a, b, c) */

/*
  mp_expt(a, b, c)

  Compute c = a ** b, that is, raise a to the b power.  Uses a
  standard iterative square-and-multiply technique.
 */

mp_err mp_expt(mp_int *a, mp_int *b, mp_int *c)
{
  mp_int   s, x;
  mp_err   res;
  mp_digit d;
  int      dig, bit;

  ARGCHK(a != NULL && b != NULL && c != NULL, MP_BADARG);

  if(mp_cmp_z(b) < 0)
    return MP_RANGE;

  if((res = mp_init(&s)) != MP_OKAY)
    return res;

  mp_set(&s, 1);

  if((res = mp_init_copy(&x, a)) != MP_OKAY)
    goto X;

  /* Loop over low-order digits in ascending order */
  for(dig = 0; dig < (USED(b) - 1); dig++) {
    d = DIGIT(b, dig);

    /* Loop over bits of each non-maximal digit */
    for(bit = 0; bit < DIGIT_BIT; bit++) {
      if(d & 1) {
	if((res = s_mp_mul(&s, &x)) != MP_OKAY) 
	  goto CLEANUP;
      }

      d >>= 1;
      
      if((res = s_mp_sqr(&x)) != MP_OKAY)
	goto CLEANUP;
    }
  }

  /* Consider now the last digit... */
  d = DIGIT(b, dig);

  while(d) {
    if(d & 1) {
      if((res = s_mp_mul(&s, &x)) != MP_OKAY)
	goto CLEANUP;
    }

    d >>= 1;

    if((res = s_mp_sqr(&x)) != MP_OKAY)
      goto CLEANUP;
  }
  
  if(mp_iseven(b))
    SIGN(&s) = SIGN(a);

  res = mp_copy(&s, c);

CLEANUP:
  mp_clear(&x);
X:
  mp_clear(&s);

  return res;

} /* end mp_expt() */

/* }}} */

/* {{{ mp_2expt(a, k) */

/* Compute a = 2^k */

mp_err mp_2expt(mp_int *a, mp_digit k)
{
  ARGCHK(a != NULL, MP_BADARG);

  return s_mp_2expt(a, k);

} /* end mp_2expt() */

/* }}} */

/* {{{ mp_mod(a, m, c) */

/*
  mp_mod(a, m, c)

  Compute c = a (mod m).  Result will always be 0 <= c < m.
 */

mp_err mp_mod(mp_int *a, mp_int *m, mp_int *c)
{
  mp_err  res;
  int     mag;

  ARGCHK(a != NULL && m != NULL && c != NULL, MP_BADARG);

  if(SIGN(m) == MP_NEG)
    return MP_RANGE;

  /*
     If |a| > m, we need to divide to get the remainder and take the
     absolute value.  

     If |a| < m, we don't need to do any division, just copy and adjust
     the sign (if a is negative).

     If |a| == m, we can simply set the result to zero.

     This order is intended to minimize the average path length of the
     comparison chain on common workloads -- the most frequent cases are
     that |a| != m, so we do those first.
   */
  if((mag = s_mp_cmp(a, m)) > 0) {
    if((res = mp_div(a, m, NULL, c)) != MP_OKAY)
      return res;
    
    if(SIGN(c) == MP_NEG) {
      if((res = mp_add(c, m, c)) != MP_OKAY)
	return res;
    }

  } else if(mag < 0) {
    if((res = mp_copy(a, c)) != MP_OKAY)
      return res;

    if(mp_cmp_z(a) < 0) {
      if((res = mp_add(c, m, c)) != MP_OKAY)
	return res;

    }
    
  } else {
    mp_zero(c);

  }

  return MP_OKAY;

} /* end mp_mod() */

/* }}} */

/* {{{ mp_mod_d(a, d, c) */

/*
  mp_mod_d(a, d, c)

  Compute c = a (mod d).  Result will always be 0 <= c < d
 */
mp_err mp_mod_d(mp_int *a, mp_digit d, mp_digit *c)
{
  mp_err   res;
  mp_digit rem;

  ARGCHK(a != NULL && c != NULL, MP_BADARG);

  if(s_mp_cmp_d(a, d) > 0) {
    if((res = mp_div_d(a, d, NULL, &rem)) != MP_OKAY)
      return res;

  } else {
    if(SIGN(a) == MP_NEG)
      rem = d - DIGIT(a, 0);
    else
      rem = DIGIT(a, 0);
  }

  if(c)
    *c = rem;

  return MP_OKAY;

} /* end mp_mod_d() */

/* }}} */

/* {{{ mp_sqrt(a, b) */

/*
  mp_sqrt(a, b)

  Compute the integer square root of a, and store the result in b.
  Uses an integer-arithmetic version of Newton's iterative linear
  approximation technique to determine this value; the result has the
  following two properties:

     b^2 <= a
     (b+1)^2 >= a

  It is a range error to pass a negative value.
 */
mp_err mp_sqrt(mp_int *a, mp_int *b)
{
  mp_int   x, t;
  mp_err   res;

  ARGCHK(a != NULL && b != NULL, MP_BADARG);

  /* Cannot take square root of a negative value */
  if(SIGN(a) == MP_NEG)
    return MP_RANGE;

  /* Special cases for zero and one, trivial     */
  if(mp_cmp_d(a, 0) == MP_EQ || mp_cmp_d(a, 1) == MP_EQ) 
    return mp_copy(a, b);
    
  /* Initialize the temporaries we'll use below  */
  if((res = mp_init_size(&t, USED(a))) != MP_OKAY)
    return res;

  /* Compute an initial guess for the iteration as a itself */
  if((res = mp_init_copy(&x, a)) != MP_OKAY)
    goto X;

s_mp_rshd(&x, (USED(&x)/2)+1);
mp_add_d(&x, 1, &x);

  for(;;) {
    /* t = (x * x) - a */
    mp_copy(&x, &t);      /* can't fail, t is big enough for original x */
    if((res = mp_sqr(&t, &t)) != MP_OKAY ||
       (res = mp_sub(&t, a, &t)) != MP_OKAY)
      goto CLEANUP;

    /* t = t / 2x       */
    s_mp_mul_2(&x);
    if((res = mp_div(&t, &x, &t, NULL)) != MP_OKAY)
      goto CLEANUP;
    s_mp_div_2(&x);

    /* Terminate the loop, if the quotient is zero */
    if(mp_cmp_z(&t) == MP_EQ)
      break;

    /* x = x - t       */
    if((res = mp_sub(&x, &t, &x)) != MP_OKAY)
      goto CLEANUP;

  }

  /* Copy result to output parameter */
  mp_sub_d(&x, 1, &x);
  s_mp_exch(&x, b);

 CLEANUP:
  mp_clear(&x);
 X:
  mp_clear(&t); 

  return res;

} /* end mp_sqrt() */

/* }}} */

/* }}} */

/*------------------------------------------------------------------------*/
/* {{{ Modular arithmetic */

#if MP_MODARITH
/* {{{ mp_addmod(a, b, m, c) */

/*
  mp_addmod(a, b, m, c)

  Compute c = (a + b) mod m
 */

mp_err mp_addmod(mp_int *a, mp_int *b, mp_int *m, mp_int *c)
{
  mp_err  res;

  ARGCHK(a != NULL && b != NULL && m != NULL && c != NULL, MP_BADARG);

  if((res = mp_add(a, b, c)) != MP_OKAY)
    return res;
  if((res = mp_mod(c, m, c)) != MP_OKAY)
    return res;

  return MP_OKAY;

}

/* }}} */

/* {{{ mp_submod(a, b, m, c) */

/*
  mp_submod(a, b, m, c)

  Compute c = (a - b) mod m
 */

mp_err mp_submod(mp_int *a, mp_int *b, mp_int *m, mp_int *c)
{
  mp_err  res;

  ARGCHK(a != NULL && b != NULL && m != NULL && c != NULL, MP_BADARG);

  if((res = mp_sub(a, b, c)) != MP_OKAY)
    return res;
  if((res = mp_mod(c, m, c)) != MP_OKAY)
    return res;

  return MP_OKAY;

}

/* }}} */

/* {{{ mp_mulmod(a, b, m, c) */

/*
  mp_mulmod(a, b, m, c)

  Compute c = (a * b) mod m
 */

mp_err mp_mulmod(mp_int *a, mp_int *b, mp_int *m, mp_int *c)
{
  mp_err  res;

  ARGCHK(a != NULL && b != NULL && m != NULL && c != NULL, MP_BADARG);

  if((res = mp_mul(a, b, c)) != MP_OKAY)
    return res;
  if((res = mp_mod(c, m, c)) != MP_OKAY)
    return res;

  return MP_OKAY;

}

/* }}} */

/* {{{ mp_sqrmod(a, m, c) */

#if MP_SQUARE
mp_err mp_sqrmod(mp_int *a, mp_int *m, mp_int *c)
{
  mp_err  res;

  ARGCHK(a != NULL && m != NULL && c != NULL, MP_BADARG);

  if((res = mp_sqr(a, c)) != MP_OKAY)
    return res;
  if((res = mp_mod(c, m, c)) != MP_OKAY)
    return res;

  return MP_OKAY;

} /* end mp_sqrmod() */
#endif

/* }}} */

/* {{{ mp_exptmod(a, b, m, c) */

/*
  mp_exptmod(a, b, m, c)

  Compute c = (a ** b) mod m.  Uses a standard square-and-multiply
  method with modular reductions at each step. (This is basically the
  same code as mp_expt(), except for the addition of the reductions)
  
  The modular reductions are done using Barrett's algorithm (see
  s_mp_reduce() below for details)
 */

mp_err mp_exptmod(mp_int *a, mp_int *b, mp_int *m, mp_int *c)
{
  mp_int   s, x, mu;
  mp_err   res;
  mp_digit d, *db = DIGITS(b);
  mp_size  ub = USED(b);
  int      dig, bit;

  ARGCHK(a != NULL && b != NULL && c != NULL, MP_BADARG);

  if(mp_cmp_z(b) < 0 || mp_cmp_z(m) <= 0)
    return MP_RANGE;

  if((res = mp_init(&s)) != MP_OKAY)
    return res;
  if((res = mp_init_copy(&x, a)) != MP_OKAY)
    goto X;
  if((res = mp_mod(&x, m, &x)) != MP_OKAY ||
     (res = mp_init(&mu)) != MP_OKAY)
    goto MU;

  mp_set(&s, 1);

  /* mu = b^2k / m */
  s_mp_add_d(&mu, 1); 
  s_mp_lshd(&mu, 2 * USED(m));
  if((res = mp_div(&mu, m, &mu, NULL)) != MP_OKAY)
    goto CLEANUP;

  /* Loop over digits of b in ascending order, except highest order */
  for(dig = 0; dig < (ub - 1); dig++) {
    d = *db++;

    /* Loop over the bits of the lower-order digits */
    for(bit = 0; bit < DIGIT_BIT; bit++) {
      if(d & 1) {
	if((res = s_mp_mul(&s, &x)) != MP_OKAY)
	  goto CLEANUP;
	if((res = s_mp_reduce(&s, m, &mu)) != MP_OKAY)
	  goto CLEANUP;
      }

      d >>= 1;

      if((res = s_mp_sqr(&x)) != MP_OKAY)
	goto CLEANUP;
      if((res = s_mp_reduce(&x, m, &mu)) != MP_OKAY)
	goto CLEANUP;
    }
  }

  /* Now do the last digit... */
  d = *db;

  while(d) {
    if(d & 1) {
      if((res = s_mp_mul(&s, &x)) != MP_OKAY)
	goto CLEANUP;
      if((res = s_mp_reduce(&s, m, &mu)) != MP_OKAY)
	goto CLEANUP;
    }

    d >>= 1;

    if((res = s_mp_sqr(&x)) != MP_OKAY)
      goto CLEANUP;
    if((res = s_mp_reduce(&x, m, &mu)) != MP_OKAY)
      goto CLEANUP;
  }

  s_mp_exch(&s, c);

 CLEANUP:
  mp_clear(&mu);
 MU:
  mp_clear(&x);
 X:
  mp_clear(&s);

  return res;

} /* end mp_exptmod() */

/* }}} */

/* {{{ mp_exptmod_d(a, d, m, c) */

mp_err mp_exptmod_d(mp_int *a, mp_digit d, mp_int *m, mp_int *c)
{
  mp_int   s, x;
  mp_err   res;

  ARGCHK(a != NULL && c != NULL, MP_BADARG);

  if((res = mp_init(&s)) != MP_OKAY)
    return res;
  if((res = mp_init_copy(&x, a)) != MP_OKAY)
    goto X;

  mp_set(&s, 1);

  while(d != 0) {
    if(d & 1) {
      if((res = s_mp_mul(&s, &x)) != MP_OKAY ||
	 (res = mp_mod(&s, m, &s)) != MP_OKAY)
	goto CLEANUP;
    }

    d /= 2;

    if((res = s_mp_sqr(&x)) != MP_OKAY ||
       (res = mp_mod(&x, m, &x)) != MP_OKAY)
      goto CLEANUP;
  }

  s_mp_exch(&s, c);

CLEANUP:
  mp_clear(&x);
X:
  mp_clear(&s);

  return res;

} /* end mp_exptmod_d() */

/* }}} */
#endif /* if MP_MODARITH */

/* }}} */

/*------------------------------------------------------------------------*/
/* {{{ Comparison functions */

/* {{{ mp_cmp_z(a) */

/*
  mp_cmp_z(a)

  Compare a <=> 0.  Returns <0 if a<0, 0 if a=0, >0 if a>0.
 */

int    mp_cmp_z(mp_int *a)
{
  if(SIGN(a) == MP_NEG)
    return MP_LT;
  else if(USED(a) == 1 && DIGIT(a, 0) == 0)
    return MP_EQ;
  else
    return MP_GT;

} /* end mp_cmp_z() */

/* }}} */

/* {{{ mp_cmp_d(a, d) */

/*
  mp_cmp_d(a, d)

  Compare a <=> d.  Returns <0 if a<d, 0 if a=d, >0 if a>d
 */

int    mp_cmp_d(mp_int *a, mp_digit d)
{
  ARGCHK(a != NULL, MP_EQ);

  if(SIGN(a) == MP_NEG)
    return MP_LT;

  return s_mp_cmp_d(a, d);

} /* end mp_cmp_d() */

/* }}} */

/* {{{ mp_cmp(a, b) */

int    mp_cmp(mp_int *a, mp_int *b)
{
  ARGCHK(a != NULL && b != NULL, MP_EQ);

  if(SIGN(a) == SIGN(b)) {
    int  mag;

    if((mag = s_mp_cmp(a, b)) == MP_EQ)
      return MP_EQ;

    if(SIGN(a) == MP_ZPOS)
      return mag;
    else
      return -mag;

  } else if(SIGN(a) == MP_ZPOS) {
    return MP_GT;
  } else {
    return MP_LT;
  }

} /* end mp_cmp() */

/* }}} */

/* {{{ mp_cmp_mag(a, b) */

/*
  mp_cmp_mag(a, b)

  Compares |a| <=> |b|, and returns an appropriate comparison result
 */

int    mp_cmp_mag(mp_int *a, mp_int *b)
{
  ARGCHK(a != NULL && b != NULL, MP_EQ);

  return s_mp_cmp(a, b);

} /* end mp_cmp_mag() */

/* }}} */

/* {{{ mp_cmp_int(a, z) */

/*
  This just converts z to an mp_int, and uses the existing comparison
  routines.  This is sort of inefficient, but it's not clear to me how
  frequently this wil get used anyway.  For small positive constants,
  you can always use mp_cmp_d(), and for zero, there is mp_cmp_z().
 */
int    mp_cmp_int(mp_int *a, long z)
{
  mp_int  tmp;
  int     out;

  ARGCHK(a != NULL, MP_EQ);
  
  mp_init(&tmp); mp_set_int(&tmp, z);
  out = mp_cmp(a, &tmp);
  mp_clear(&tmp);

  return out;

} /* end mp_cmp_int() */

/* }}} */

/* {{{ mp_isodd(a) */

/*
  mp_isodd(a)

  Returns a true (non-zero) value if a is odd, false (zero) otherwise.
 */
int    mp_isodd(mp_int *a)
{
  ARGCHK(a != NULL, 0);

  return (DIGIT(a, 0) & 1);

} /* end mp_isodd() */

/* }}} */

/* {{{ mp_iseven(a) */

int    mp_iseven(mp_int *a)
{
  return !mp_isodd(a);

} /* end mp_iseven() */

/* }}} */

/* }}} */

/*------------------------------------------------------------------------*/
/* {{{ Number theoretic functions */

#if MP_NUMTH
/* {{{ mp_gcd(a, b, c) */

/*
  Like the old mp_gcd() function, except computes the GCD using the
  binary algorithm due to Josef Stein in 1961 (via Knuth).
 */
mp_err mp_gcd(mp_int *a, mp_int *b, mp_int *c)
{
  mp_err   res;
  mp_int   u, v, t;
  mp_size  k = 0;

  ARGCHK(a != NULL && b != NULL && c != NULL, MP_BADARG);

  if(mp_cmp_z(a) == MP_EQ && mp_cmp_z(b) == MP_EQ)
      return MP_RANGE;
  if(mp_cmp_z(a) == MP_EQ) {
    return mp_copy(b, c);
  } else if(mp_cmp_z(b) == MP_EQ) {
    return mp_copy(a, c);
  }

  if((res = mp_init(&t)) != MP_OKAY)
    return res;
  if((res = mp_init_copy(&u, a)) != MP_OKAY)
    goto U;
  if((res = mp_init_copy(&v, b)) != MP_OKAY)
    goto V;

  SIGN(&u) = MP_ZPOS;
  SIGN(&v) = MP_ZPOS;

  /* Divide out common factors of 2 until at least 1 of a, b is even */
  while(mp_iseven(&u) && mp_iseven(&v)) {
    s_mp_div_2(&u);
    s_mp_div_2(&v);
    ++k;
  }

  /* Initialize t */
  if(mp_isodd(&u)) {
    if((res = mp_copy(&v, &t)) != MP_OKAY)
      goto CLEANUP;
    
    /* t = -v */
    if(SIGN(&v) == MP_ZPOS)
      SIGN(&t) = MP_NEG;
    else
      SIGN(&t) = MP_ZPOS;
    
  } else {
    if((res = mp_copy(&u, &t)) != MP_OKAY)
      goto CLEANUP;

  }

  for(;;) {
    while(mp_iseven(&t)) {
      s_mp_div_2(&t);
    }

    if(mp_cmp_z(&t) == MP_GT) {
      if((res = mp_copy(&t, &u)) != MP_OKAY)
	goto CLEANUP;

    } else {
      if((res = mp_copy(&t, &v)) != MP_OKAY)
	goto CLEANUP;

      /* v = -t */
      if(SIGN(&t) == MP_ZPOS)
	SIGN(&v) = MP_NEG;
      else
	SIGN(&v) = MP_ZPOS;
    }

    if((res = mp_sub(&u, &v, &t)) != MP_OKAY)
      goto CLEANUP;

    if(s_mp_cmp_d(&t, 0) == MP_EQ)
      break;
  }

  s_mp_2expt(&v, k);       /* v = 2^k   */
  res = mp_mul(&u, &v, c); /* c = u * v */

 CLEANUP:
  mp_clear(&v);
 V:
  mp_clear(&u);
 U:
  mp_clear(&t);

  return res;

} /* end mp_bgcd() */

/* }}} */

/* {{{ mp_lcm(a, b, c) */

/* We compute the least common multiple using the rule:

   ab = [a, b](a, b)

   ... by computing the product, and dividing out the gcd.
 */

mp_err mp_lcm(mp_int *a, mp_int *b, mp_int *c)
{
  mp_int  gcd, prod;
  mp_err  res;

  ARGCHK(a != NULL && b != NULL && c != NULL, MP_BADARG);

  /* Set up temporaries */
  if((res = mp_init(&gcd)) != MP_OKAY)
    return res;
  if((res = mp_init(&prod)) != MP_OKAY)
    goto GCD;

  if((res = mp_mul(a, b, &prod)) != MP_OKAY)
    goto CLEANUP;
  if((res = mp_gcd(a, b, &gcd)) != MP_OKAY)
    goto CLEANUP;

  res = mp_div(&prod, &gcd, c, NULL);

 CLEANUP:
  mp_clear(&prod);
 GCD:
  mp_clear(&gcd);

  return res;

} /* end mp_lcm() */

/* }}} */

/* {{{ mp_xgcd(a, b, g, x, y) */

/*
  mp_xgcd(a, b, g, x, y)

  Compute g = (a, b) and values x and y satisfying Bezout's identity
  (that is, ax + by = g).  This uses the extended binary GCD algorithm
  based on the Stein algorithm used for mp_gcd()
 */

mp_err mp_xgcd(mp_int *a, mp_int *b, mp_int *g, mp_int *x, mp_int *y)
{
  mp_int   gx, xc, yc, u, v, A, B, C, D;
  mp_int  *clean[9];
  mp_err   res;
  int      last = -1;

  if(mp_cmp_z(b) == 0)
    return MP_RANGE;

  /* Initialize all these variables we need */
  if((res = mp_init(&u)) != MP_OKAY) goto CLEANUP;
  clean[++last] = &u;
  if((res = mp_init(&v)) != MP_OKAY) goto CLEANUP;
  clean[++last] = &v;
  if((res = mp_init(&gx)) != MP_OKAY) goto CLEANUP;
  clean[++last] = &gx;
  if((res = mp_init(&A)) != MP_OKAY) goto CLEANUP;
  clean[++last] = &A;
  if((res = mp_init(&B)) != MP_OKAY) goto CLEANUP;
  clean[++last] = &B;
  if((res = mp_init(&C)) != MP_OKAY) goto CLEANUP;
  clean[++last] = &C;
  if((res = mp_init(&D)) != MP_OKAY) goto CLEANUP;
  clean[++last] = &D;
  if((res = mp_init_copy(&xc, a)) != MP_OKAY) goto CLEANUP;
  clean[++last] = &xc;
  mp_abs(&xc, &xc);
  if((res = mp_init_copy(&yc, b)) != MP_OKAY) goto CLEANUP;
  clean[++last] = &yc;
  mp_abs(&yc, &yc);

  mp_set(&gx, 1);

  /* Divide by two until at least one of them is even */
  while(mp_iseven(&xc) && mp_iseven(&yc)) {
    s_mp_div_2(&xc);
    s_mp_div_2(&yc);
    if((res = s_mp_mul_2(&gx)) != MP_OKAY)
      goto CLEANUP;
  }

  mp_copy(&xc, &u);
  mp_copy(&yc, &v);
  mp_set(&A, 1); mp_set(&D, 1);

  /* Loop through binary GCD algorithm */
  for(;;) {
    while(mp_iseven(&u)) {
      s_mp_div_2(&u);

      if(mp_iseven(&A) && mp_iseven(&B)) {
	s_mp_div_2(&A); s_mp_div_2(&B);
      } else {
	if((res = mp_add(&A, &yc, &A)) != MP_OKAY) goto CLEANUP;
	s_mp_div_2(&A);
	if((res = mp_sub(&B, &xc, &B)) != MP_OKAY) goto CLEANUP;
	s_mp_div_2(&B);
      }
    }

    while(mp_iseven(&v)) {
      s_mp_div_2(&v);

      if(mp_iseven(&C) && mp_iseven(&D)) {
	s_mp_div_2(&C); s_mp_div_2(&D);
      } else {
	if((res = mp_add(&C, &yc, &C)) != MP_OKAY) goto CLEANUP;
	s_mp_div_2(&C);
	if((res = mp_sub(&D, &xc, &D)) != MP_OKAY) goto CLEANUP;
	s_mp_div_2(&D);
      }
    }

    if(mp_cmp(&u, &v) >= 0) {
      if((res = mp_sub(&u, &v, &u)) != MP_OKAY) goto CLEANUP;
      if((res = mp_sub(&A, &C, &A)) != MP_OKAY) goto CLEANUP;
      if((res = mp_sub(&B, &D, &B)) != MP_OKAY) goto CLEANUP;

    } else {
      if((res = mp_sub(&v, &u, &v)) != MP_OKAY) goto CLEANUP;
      if((res = mp_sub(&C, &A, &C)) != MP_OKAY) goto CLEANUP;
      if((res = mp_sub(&D, &B, &D)) != MP_OKAY) goto CLEANUP;

    }

    /* If we're done, copy results to output */
    if(mp_cmp_z(&u) == 0) {
      if(x)
	if((res = mp_copy(&C, x)) != MP_OKAY) goto CLEANUP;

      if(y)
	if((res = mp_copy(&D, y)) != MP_OKAY) goto CLEANUP;
      
      if(g)
	if((res = mp_mul(&gx, &v, g)) != MP_OKAY) goto CLEANUP;

      break;
    }
  }

 CLEANUP:
  while(last >= 0)
    mp_clear(clean[last--]);

  return res;

} /* end mp_xgcd() */

/* }}} */

/* {{{ mp_invmod(a, m, c) */

/*
  mp_invmod(a, m, c)

  Compute c = a^-1 (mod m), if there is an inverse for a (mod m).
  This is equivalent to the question of whether (a, m) = 1.  If not,
  MP_UNDEF is returned, and there is no inverse.
 */

mp_err mp_invmod(mp_int *a, mp_int *m, mp_int *c)
{
  mp_int  g, x;
  mp_err  res;

  ARGCHK(a && m && c, MP_BADARG);

  if(mp_cmp_z(a) == 0 || mp_cmp_z(m) == 0)
    return MP_RANGE;

  if((res = mp_init(&g)) != MP_OKAY)
    return res;
  if((res = mp_init(&x)) != MP_OKAY)
    goto X;

  if((res = mp_xgcd(a, m, &g, &x, NULL)) != MP_OKAY)
    goto CLEANUP;

  if(mp_cmp_d(&g, 1) != MP_EQ) {
    res = MP_UNDEF;
    goto CLEANUP;
  }

  res = mp_mod(&x, m, c);
  SIGN(c) = SIGN(a);

CLEANUP:
  mp_clear(&x);
X:
  mp_clear(&g);

  return res;

} /* end mp_invmod() */

/* }}} */
#endif /* if MP_NUMTH */

/* }}} */

/*------------------------------------------------------------------------*/
/* {{{ mp_print(mp, ofp) */

#if MP_IOFUNC
/*
  mp_print(mp, ofp)

  Print a textual representation of the given mp_int on the output
  stream 'ofp'.  Output is generated using the internal radix.
 */

void   mp_print(mp_int *mp, FILE *ofp)
{
  int   ix;

  if(mp == NULL || ofp == NULL)
    return;

  fputc((SIGN(mp) == MP_NEG) ? '-' : '+', ofp);

  for(ix = USED(mp) - 1; ix >= 0; ix--) {
    fprintf(ofp, DIGIT_FMT, DIGIT(mp, ix));
  }

} /* end mp_print() */

#endif /* if MP_IOFUNC */

/* }}} */

/*------------------------------------------------------------------------*/
/* {{{ More I/O Functions */

/* {{{ mp_read_signed_bin(mp, str, len) */

/* 
   mp_read_signed_bin(mp, str, len)

   Read in a raw value (base 256) into the given mp_int
 */

mp_err  mp_read_signed_bin(mp_int *mp, unsigned char *str, int len)
{
  mp_err         res;

  ARGCHK(mp != NULL && str != NULL && len > 0, MP_BADARG);

  if((res = mp_read_unsigned_bin(mp, str + 1, len - 1)) == MP_OKAY) {
    /* Get sign from first byte */
    if(str[0])
      SIGN(mp) = MP_NEG;
    else
      SIGN(mp) = MP_ZPOS;
  }

  return res;

} /* end mp_read_signed_bin() */

/* }}} */

/* {{{ mp_signed_bin_size(mp) */

int    mp_signed_bin_size(mp_int *mp)
{
  ARGCHK(mp != NULL, 0);

  return mp_unsigned_bin_size(mp) + 1;

} /* end mp_signed_bin_size() */

/* }}} */

/* {{{ mp_to_signed_bin(mp, str) */

mp_err mp_to_signed_bin(mp_int *mp, unsigned char *str)
{
  ARGCHK(mp != NULL && str != NULL, MP_BADARG);

  /* Caller responsible for allocating enough memory (use mp_raw_size(mp)) */
  str[0] = (char)SIGN(mp);

  return mp_to_unsigned_bin(mp, str + 1);

} /* end mp_to_signed_bin() */

/* }}} */

/* {{{ mp_read_unsigned_bin(mp, str, len) */

/*
  mp_read_unsigned_bin(mp, str, len)

  Read in an unsigned value (base 256) into the given mp_int
 */

mp_err  mp_read_unsigned_bin(mp_int *mp, unsigned char *str, int len)
{
  int     ix;
  mp_err  res;

  ARGCHK(mp != NULL && str != NULL && len > 0, MP_BADARG);

  mp_zero(mp);

  for(ix = 0; ix < len; ix++) {
    if((res = s_mp_mul_2d(mp, CHAR_BIT)) != MP_OKAY)
      return res;

    if((res = mp_add_d(mp, str[ix], mp)) != MP_OKAY)
      return res;
  }
  
  return MP_OKAY;
  
} /* end mp_read_unsigned_bin() */

/* }}} */

/* {{{ mp_unsigned_bin_size(mp) */

int     mp_unsigned_bin_size(mp_int *mp) 
{
  mp_digit   topdig;
  int        count;

  ARGCHK(mp != NULL, 0);

  /* Special case for the value zero */
  if(USED(mp) == 1 && DIGIT(mp, 0) == 0)
    return 1;

  count = (USED(mp) - 1) * sizeof(mp_digit);
  topdig = DIGIT(mp, USED(mp) - 1);

  while(topdig != 0) {
    ++count;
    topdig >>= CHAR_BIT;
  }

  return count;

} /* end mp_unsigned_bin_size() */

/* }}} */

/* {{{ mp_to_unsigned_bin(mp, str) */

mp_err mp_to_unsigned_bin(mp_int *mp, unsigned char *str)
{
  mp_digit      *dp, *end, d;
  unsigned char *spos;

  ARGCHK(mp != NULL && str != NULL, MP_BADARG);

  dp = DIGITS(mp);
  end = dp + USED(mp) - 1;
  spos = str;

  /* Special case for zero, quick test */
  if(dp == end && *dp == 0) {
    *str = '\0';
    return MP_OKAY;
  }

  /* Generate digits in reverse order */
  while(dp < end) {
    int      ix;

    d = *dp;
    for(ix = 0; ix < sizeof(mp_digit); ++ix) {
      *spos = d & UCHAR_MAX;
      d >>= CHAR_BIT;
      ++spos;
    }

    ++dp;
  }

  /* Now handle last digit specially, high order zeroes are not written */
  d = *end;
  while(d != 0) {
    *spos = d & UCHAR_MAX;
    d >>= CHAR_BIT;
    ++spos;
  }

  /* Reverse everything to get digits in the correct order */
  while(--spos > str) {
    unsigned char t = *str;
    *str = *spos;
    *spos = t;

    ++str;
  }

  return MP_OKAY;

} /* end mp_to_unsigned_bin() */

/* }}} */

/* {{{ mp_count_bits(mp) */

int    mp_count_bits(mp_int *mp)
{
  int      len;
  mp_digit d;

  ARGCHK(mp != NULL, MP_BADARG);

  len = DIGIT_BIT * (USED(mp) - 1);
  d = DIGIT(mp, USED(mp) - 1);

  while(d != 0) {
    ++len;
    d >>= 1;
  }

  return len;
  
} /* end mp_count_bits() */

/* }}} */

/* {{{ mp_read_radix(mp, str, radix) */

/*
  mp_read_radix(mp, str, radix)

  Read an integer from the given string, and set mp to the resulting
  value.  The input is presumed to be in base 10.  Leading non-digit
  characters are ignored, and the function reads until a non-digit
  character or the end of the string.
 */

mp_err  mp_read_radix(mp_int *mp, unsigned char *str, int radix)
{
  int     ix = 0, val = 0;
  mp_err  res;
  mp_sign sig = MP_ZPOS;

  ARGCHK(mp != NULL && str != NULL && radix >= 2 && radix <= MAX_RADIX, 
	 MP_BADARG);

  mp_zero(mp);

  /* Skip leading non-digit characters until a digit or '-' or '+' */
  while(str[ix] && 
	(s_mp_tovalue(str[ix], radix) < 0) && 
	str[ix] != '-' &&
	str[ix] != '+') {
    ++ix;
  }

  if(str[ix] == '-') {
    sig = MP_NEG;
    ++ix;
  } else if(str[ix] == '+') {
    sig = MP_ZPOS; /* this is the default anyway... */
    ++ix;
  }

  while((val = s_mp_tovalue(str[ix], radix)) >= 0) {
    if((res = s_mp_mul_d(mp, radix)) != MP_OKAY)
      return res;
    if((res = s_mp_add_d(mp, val)) != MP_OKAY)
      return res;
    ++ix;
  }

  if(s_mp_cmp_d(mp, 0) == MP_EQ)
    SIGN(mp) = MP_ZPOS;
  else
    SIGN(mp) = sig;

  return MP_OKAY;

} /* end mp_read_radix() */

/* }}} */

/* {{{ mp_radix_size(mp, radix) */

int    mp_radix_size(mp_int *mp, int radix)
{
  int  len;
  ARGCHK(mp != NULL, 0);

  len = s_mp_outlen(mp_count_bits(mp), radix) + 1; /* for NUL terminator */

  if(mp_cmp_z(mp) < 0)
    ++len; /* for sign */

  return len;

} /* end mp_radix_size() */

/* }}} */

/* {{{ mp_value_radix_size(num, qty, radix) */

/* num = number of digits
   qty = number of bits per digit
   radix = target base
   
   Return the number of digits in the specified radix that would be
   needed to express 'num' digits of 'qty' bits each.
 */
int    mp_value_radix_size(int num, int qty, int radix)
{
  ARGCHK(num >= 0 && qty > 0 && radix >= 2 && radix <= MAX_RADIX, 0);

  return s_mp_outlen(num * qty, radix);

} /* end mp_value_radix_size() */

/* }}} */

/* {{{ mp_toradix(mp, str, radix) */

mp_err mp_toradix(mp_int *mp, unsigned char *str, int radix)
{
  int  ix, pos = 0;

  ARGCHK(mp != NULL && str != NULL, MP_BADARG);
  ARGCHK(radix > 1 && radix <= MAX_RADIX, MP_RANGE);

  if(mp_cmp_z(mp) == MP_EQ) {
    str[0] = '0';
    str[1] = '\0';
  } else {
    mp_err   res;
    mp_int   tmp;
    mp_sign  sgn;
    mp_digit rem, rdx = (mp_digit)radix;
    char     ch;

    if((res = mp_init_copy(&tmp, mp)) != MP_OKAY)
      return res;

    /* Save sign for later, and take absolute value */
    sgn = SIGN(&tmp); SIGN(&tmp) = MP_ZPOS;

    /* Generate output digits in reverse order      */
    while(mp_cmp_z(&tmp) != 0) {
      if((res = s_mp_div_d(&tmp, rdx, &rem)) != MP_OKAY) {
	mp_clear(&tmp);
	return res;
      }

      /* Generate digits, use capital letters */
      ch = s_mp_todigit(rem, radix, 0);

      str[pos++] = ch;
    }

    /* Add - sign if original value was negative */
    if(sgn == MP_NEG)
      str[pos++] = '-';

    /* Add trailing NUL to end the string        */
    str[pos--] = '\0';

    /* Reverse the digits and sign indicator     */
    ix = 0;
    while(ix < pos) {
      char tmp = str[ix];

      str[ix] = str[pos];
      str[pos] = tmp;
      ++ix;
      --pos;
    }
    
    mp_clear(&tmp);
  }

  return MP_OKAY;

} /* end mp_toradix() */

/* }}} */

/* {{{ mp_char2value(ch, r) */

int    mp_char2value(char ch, int r)
{
  return s_mp_tovalue(ch, r);

} /* end mp_tovalue() */

/* }}} */

/* }}} */

/* {{{ mp_strerror(ec) */

/*
  mp_strerror(ec)

  Return a string describing the meaning of error code 'ec'.  The
  string returned is allocated in static memory, so the caller should
  not attempt to modify or free the memory associated with this
  string.
 */
const char  *mp_strerror(mp_err ec)
{
  int   aec = (ec < 0) ? -ec : ec;

  /* Code values are negative, so the senses of these comparisons
     are accurate */
  if(ec < MP_LAST_CODE || ec > MP_OKAY) {
    return mp_err_string[0];  /* unknown error code */
  } else {
    return mp_err_string[aec + 1];
  }

} /* end mp_strerror() */

/* }}} */

/*========================================================================*/
/*------------------------------------------------------------------------*/
/* Static function definitions (internal use only)                        */

/* {{{ Memory management */

/* {{{ s_mp_grow(mp, min) */

/* Make sure there are at least 'min' digits allocated to mp              */
mp_err   s_mp_grow(mp_int *mp, mp_size min)
{
  if(min > ALLOC(mp)) {
    mp_digit   *tmp;

    /* Set min to next nearest default precision block size */
    min = ((min + (s_mp_defprec - 1)) / s_mp_defprec) * s_mp_defprec;

    if((tmp = s_mp_alloc(min, sizeof(mp_digit))) == NULL)
      return MP_MEM;

    s_mp_copy(DIGITS(mp), tmp, USED(mp));

#if MP_CRYPTO
    s_mp_setz(DIGITS(mp), ALLOC(mp));
#endif
    s_mp_free(DIGITS(mp));
    DIGITS(mp) = tmp;
    ALLOC(mp) = min;
  }

  return MP_OKAY;

} /* end s_mp_grow() */

/* }}} */

/* {{{ s_mp_pad(mp, min) */

/* Make sure the used size of mp is at least 'min', growing if needed     */
mp_err   s_mp_pad(mp_int *mp, mp_size min)
{
  if(min > USED(mp)) {
    mp_err  res;

    /* Make sure there is room to increase precision  */
    if(min > ALLOC(mp) && (res = s_mp_grow(mp, min)) != MP_OKAY)
      return res;

    /* Increase precision; should already be 0-filled */
    USED(mp) = min;
  }

  return MP_OKAY;

} /* end s_mp_pad() */

/* }}} */

/* {{{ s_mp_setz(dp, count) */

#if MP_MACRO == 0
/* Set 'count' digits pointed to by dp to be zeroes                       */
void s_mp_setz(mp_digit *dp, mp_size count)
{
#if MP_MEMSET == 0
  int  ix;

  for(ix = 0; ix < count; ix++)
    dp[ix] = 0;
#else
  memset(dp, 0, count * sizeof(mp_digit));
#endif

} /* end s_mp_setz() */
#endif

/* }}} */

/* {{{ s_mp_copy(sp, dp, count) */

#if MP_MACRO == 0
/* Copy 'count' digits from sp to dp                                      */
void s_mp_copy(mp_digit *sp, mp_digit *dp, mp_size count)
{
#if MP_MEMCPY == 0
  int  ix;

  for(ix = 0; ix < count; ix++)
    dp[ix] = sp[ix];
#else
  memcpy(dp, sp, count * sizeof(mp_digit));
#endif

} /* end s_mp_copy() */
#endif

/* }}} */

/* {{{ s_mp_alloc(nb, ni) */

#if MP_MACRO == 0
/* Allocate ni records of nb bytes each, and return a pointer to that     */
void    *s_mp_alloc(size_t nb, size_t ni)
{
  return calloc(nb, ni);

} /* end s_mp_alloc() */
#endif

/* }}} */

/* {{{ s_mp_free(ptr) */

#if MP_MACRO == 0
/* Free the memory pointed to by ptr                                      */
void     s_mp_free(void *ptr)
{
  if(ptr)
    free(ptr);

} /* end s_mp_free() */
#endif

/* }}} */

/* {{{ s_mp_clamp(mp) */

/* Remove leading zeroes from the given value                             */
void     s_mp_clamp(mp_int *mp)
{
  mp_size   du = USED(mp);
  mp_digit *zp = DIGITS(mp) + du - 1;

  while(du > 1 && !*zp--)
    --du;

  USED(mp) = du;

} /* end s_mp_clamp() */


/* }}} */

/* {{{ s_mp_exch(a, b) */

/* Exchange the data for a and b; (b, a) = (a, b)                         */
void     s_mp_exch(mp_int *a, mp_int *b)
{
  mp_int   tmp;

  tmp = *a;
  *a = *b;
  *b = tmp;

} /* end s_mp_exch() */

/* }}} */

/* }}} */

/* {{{ Arithmetic helpers */

/* {{{ s_mp_lshd(mp, p) */

/* 
   Shift mp leftward by p digits, growing if needed, and zero-filling
   the in-shifted digits at the right end.  This is a convenient
   alternative to multiplication by powers of the radix
 */   

mp_err   s_mp_lshd(mp_int *mp, mp_size p)
{
  mp_err   res;
  mp_size  pos;
  mp_digit *dp;
  int     ix;

  if(p == 0)
    return MP_OKAY;

  if((res = s_mp_pad(mp, USED(mp) + p)) != MP_OKAY)
    return res;

  pos = USED(mp) - 1;
  dp = DIGITS(mp);

  /* Shift all the significant figures over as needed */
  for(ix = pos - p; ix >= 0; ix--) 
    dp[ix + p] = dp[ix];

  /* Fill the bottom digits with zeroes */
  for(ix = 0; ix < p; ix++)
    dp[ix] = 0;

  return MP_OKAY;

} /* end s_mp_lshd() */

/* }}} */

/* {{{ s_mp_rshd(mp, p) */

/* 
   Shift mp rightward by p digits.  Maintains the invariant that
   digits above the precision are all zero.  Digits shifted off the
   end are lost.  Cannot fail.
 */

void     s_mp_rshd(mp_int *mp, mp_size p)
{
  mp_size  ix;
  mp_digit *dp;

  if(p == 0)
    return;

  /* Shortcut when all digits are to be shifted off */
  if(p >= USED(mp)) {
    s_mp_setz(DIGITS(mp), ALLOC(mp));
    USED(mp) = 1;
    SIGN(mp) = MP_ZPOS;
    return;
  }

  /* Shift all the significant figures over as needed */
  dp = DIGITS(mp);
  for(ix = p; ix < USED(mp); ix++)
    dp[ix - p] = dp[ix];

  /* Fill the top digits with zeroes */
  ix -= p;
  while(ix < USED(mp))
    dp[ix++] = 0;

  /* Strip off any leading zeroes    */
  s_mp_clamp(mp);

} /* end s_mp_rshd() */

/* }}} */

/* {{{ s_mp_div_2(mp) */

/* Divide by two -- take advantage of radix properties to do it fast      */
void     s_mp_div_2(mp_int *mp)
{
  s_mp_div_2d(mp, 1);

} /* end s_mp_div_2() */

/* }}} */

/* {{{ s_mp_mul_2(mp) */

mp_err s_mp_mul_2(mp_int *mp)
{
  int      ix;
  mp_digit kin = 0, kout, *dp = DIGITS(mp);
  mp_err   res;

  /* Shift digits leftward by 1 bit */
  for(ix = 0; ix < USED(mp); ix++) {
    kout = (dp[ix] >> (DIGIT_BIT - 1)) & 1;
    dp[ix] = (dp[ix] << 1) | kin;

    kin = kout;
  }

  /* Deal with rollover from last digit */
  if(kin) {
    if(ix >= ALLOC(mp)) {
      if((res = s_mp_grow(mp, ALLOC(mp) + 1)) != MP_OKAY)
	return res;
      dp = DIGITS(mp);
    }

    dp[ix] = kin;
    USED(mp) += 1;
  }

  return MP_OKAY;

} /* end s_mp_mul_2() */

/* }}} */

/* {{{ s_mp_mod_2d(mp, d) */

/*
  Remainder the integer by 2^d, where d is a number of bits.  This
  amounts to a bitwise AND of the value, and does not require the full
  division code
 */
void     s_mp_mod_2d(mp_int *mp, mp_digit d)
{
  unsigned int  ndig = (d / DIGIT_BIT), nbit = (d % DIGIT_BIT);
  unsigned int  ix;
  mp_digit      dmask, *dp = DIGITS(mp);

  if(ndig >= USED(mp))
    return;

  /* Flush all the bits above 2^d in its digit */
  dmask = (1 << nbit) - 1;
  dp[ndig] &= dmask;

  /* Flush all digits above the one with 2^d in it */
  for(ix = ndig + 1; ix < USED(mp); ix++)
    dp[ix] = 0;

  s_mp_clamp(mp);

} /* end s_mp_mod_2d() */

/* }}} */

/* {{{ s_mp_mul_2d(mp, d) */

/*
  Multiply by the integer 2^d, where d is a number of bits.  This
  amounts to a bitwise shift of the value, and does not require the
  full multiplication code.
 */
mp_err    s_mp_mul_2d(mp_int *mp, mp_digit d)
{
  mp_err   res;
  mp_digit save, next, mask, *dp;
  mp_size  used;
  int      ix;

  if((res = s_mp_lshd(mp, d / DIGIT_BIT)) != MP_OKAY)
    return res;

  dp = DIGITS(mp); used = USED(mp);
  d %= DIGIT_BIT;

  mask = (1 << d) - 1;

  /* If the shift requires another digit, make sure we've got one to
     work with */
  if((dp[used - 1] >> (DIGIT_BIT - d)) & mask) {
    if((res = s_mp_grow(mp, used + 1)) != MP_OKAY)
      return res;
    dp = DIGITS(mp);
  }

  /* Do the shifting... */
  save = 0;
  for(ix = 0; ix < used; ix++) {
    next = (dp[ix] >> (DIGIT_BIT - d)) & mask;
    dp[ix] = (dp[ix] << d) | save;
    save = next;
  }

  /* If, at this point, we have a nonzero carryout into the next
     digit, we'll increase the size by one digit, and store it...
   */
  if(save) {
    dp[used] = save;
    USED(mp) += 1;
  }

  s_mp_clamp(mp);
  return MP_OKAY;

} /* end s_mp_mul_2d() */

/* }}} */

/* {{{ s_mp_div_2d(mp, d) */

/*
  Divide the integer by 2^d, where d is a number of bits.  This
  amounts to a bitwise shift of the value, and does not require the
  full division code (used in Barrett reduction, see below)
 */
void     s_mp_div_2d(mp_int *mp, mp_digit d)
{
  int       ix;
  mp_digit  save, next, mask, *dp = DIGITS(mp);

  s_mp_rshd(mp, d / DIGIT_BIT);
  d %= DIGIT_BIT;

  mask = (1 << d) - 1;

  save = 0;
  for(ix = USED(mp) - 1; ix >= 0; ix--) {
    next = dp[ix] & mask;
    dp[ix] = (dp[ix] >> d) | (save << (DIGIT_BIT - d));
    save = next;
  }

  s_mp_clamp(mp);

} /* end s_mp_div_2d() */

/* }}} */

/* {{{ s_mp_norm(a, b) */

/*
  s_mp_norm(a, b)

  Normalize a and b for division, where b is the divisor.  In order
  that we might make good guesses for quotient digits, we want the
  leading digit of b to be at least half the radix, which we
  accomplish by multiplying a and b by a constant.  This constant is
  returned (so that it can be divided back out of the remainder at the
  end of the division process).

  We multiply by the smallest power of 2 that gives us a leading digit
  at least half the radix.  By choosing a power of 2, we simplify the 
  multiplication and division steps to simple shifts.
 */
mp_digit s_mp_norm(mp_int *a, mp_int *b)
{
  mp_digit  t, d = 0;

  t = DIGIT(b, USED(b) - 1);
  while(t < (RADIX / 2)) {
    t <<= 1;
    ++d;
  }
    
  if(d != 0) {
    s_mp_mul_2d(a, d);
    s_mp_mul_2d(b, d);
  }

  return d;

} /* end s_mp_norm() */

/* }}} */

/* }}} */

/* {{{ Primitive digit arithmetic */

/* {{{ s_mp_add_d(mp, d) */

/* Add d to |mp| in place                                                 */
mp_err   s_mp_add_d(mp_int *mp, mp_digit d)    /* unsigned digit addition */
{
  mp_word   w, k = 0;
  mp_size   ix = 1, used = USED(mp);
  mp_digit *dp = DIGITS(mp);

  w = dp[0] + d;
  dp[0] = ACCUM(w);
  k = CARRYOUT(w);

  while(ix < used && k) {
    w = dp[ix] + k;
    dp[ix] = ACCUM(w);
    k = CARRYOUT(w);
    ++ix;
  }

  if(k != 0) {
    mp_err  res;

    if((res = s_mp_pad(mp, USED(mp) + 1)) != MP_OKAY)
      return res;

    DIGIT(mp, ix) = k;
  }

  return MP_OKAY;

} /* end s_mp_add_d() */

/* }}} */

/* {{{ s_mp_sub_d(mp, d) */

/* Subtract d from |mp| in place, assumes |mp| > d                        */
mp_err   s_mp_sub_d(mp_int *mp, mp_digit d)    /* unsigned digit subtract */
{
  mp_word   w, b = 0;
  mp_size   ix = 1, used = USED(mp);
  mp_digit *dp = DIGITS(mp);

  /* Compute initial subtraction    */
  w = (RADIX + dp[0]) - d;
  b = CARRYOUT(w) ? 0 : 1;
  dp[0] = ACCUM(w);

  /* Propagate borrows leftward     */
  while(b && ix < used) {
    w = (RADIX + dp[ix]) - b;
    b = CARRYOUT(w) ? 0 : 1;
    dp[ix] = ACCUM(w);
    ++ix;
  }

  /* Remove leading zeroes          */
  s_mp_clamp(mp);

  /* If we have a borrow out, it's a violation of the input invariant */
  if(b)
    return MP_RANGE;
  else
    return MP_OKAY;

} /* end s_mp_sub_d() */

/* }}} */

/* {{{ s_mp_mul_d(a, d) */

/* Compute a = a * d, single digit multiplication                         */
mp_err   s_mp_mul_d(mp_int *a, mp_digit d)
{
  mp_word w, k = 0;
  mp_size ix, max;
  mp_err  res;
  mp_digit *dp = DIGITS(a);

  /*
    Single-digit multiplication will increase the precision of the
    output by at most one digit.  However, we can detect when this
    will happen -- if the high-order digit of a, times d, gives a
    two-digit result, then the precision of the result will increase;
    otherwise it won't.  We use this fact to avoid calling s_mp_pad()
    unless absolutely necessary.
   */
  max = USED(a);
  w = dp[max - 1] * d;
  if(CARRYOUT(w) != 0) {
    if((res = s_mp_pad(a, max + 1)) != MP_OKAY)
      return res;
    dp = DIGITS(a);
  }

  for(ix = 0; ix < max; ix++) {
    w = (dp[ix] * d) + k;
    dp[ix] = ACCUM(w);
    k = CARRYOUT(w);
  }

  /* If there is a precision increase, take care of it here; the above
     test guarantees we have enough storage to do this safely.
   */
  if(k) {
    dp[max] = k; 
    USED(a) = max + 1;
  }

  s_mp_clamp(a);

  return MP_OKAY;
  
} /* end s_mp_mul_d() */

/* }}} */

/* {{{ s_mp_div_d(mp, d, r) */

/*
  s_mp_div_d(mp, d, r)

  Compute the quotient mp = mp / d and remainder r = mp mod d, for a
  single digit d.  If r is null, the remainder will be discarded.
 */

mp_err   s_mp_div_d(mp_int *mp, mp_digit d, mp_digit *r)
{
  mp_word   w = 0, t;
  mp_int    quot;
  mp_err    res;
  mp_digit *dp = DIGITS(mp), *qp;
  int       ix;

  if(d == 0)
    return MP_RANGE;

  /* Make room for the quotient */
  if((res = mp_init_size(&quot, USED(mp))) != MP_OKAY)
    return res;

  USED(&quot) = USED(mp); /* so clamping will work below */
  qp = DIGITS(&quot);

  /* Divide without subtraction */
  for(ix = USED(mp) - 1; ix >= 0; ix--) {
    w = (w << DIGIT_BIT) | dp[ix];

    if(w >= d) {
      t = w / d;
      w = w % d;
    } else {
      t = 0;
    }

    qp[ix] = t;
  }

  /* Deliver the remainder, if desired */
  if(r)
    *r = w;

  s_mp_clamp(&quot);
  mp_exch(&quot, mp);
  mp_clear(&quot);

  return MP_OKAY;

} /* end s_mp_div_d() */

/* }}} */

/* }}} */

/* {{{ Primitive full arithmetic */

/* {{{ s_mp_add(a, b) */

/* Compute a = |a| + |b|                                                  */
mp_err   s_mp_add(mp_int *a, mp_int *b)        /* magnitude addition      */
{
  mp_word   w = 0;
  mp_digit *pa, *pb;
  mp_size   ix, used = USED(b);
  mp_err    res;

  /* Make sure a has enough precision for the output value */
  if((used > USED(a)) && (res = s_mp_pad(a, used)) != MP_OKAY)
    return res;

  /*
    Add up all digits up to the precision of b.  If b had initially
    the same precision as a, or greater, we took care of it by the
    padding step above, so there is no problem.  If b had initially
    less precision, we'll have to make sure the carry out is duly
    propagated upward among the higher-order digits of the sum.
   */
  pa = DIGITS(a);
  pb = DIGITS(b);
  for(ix = 0; ix < used; ++ix) {
    w += *pa + *pb++;
    *pa++ = ACCUM(w);
    w = CARRYOUT(w);
  }

  /* If we run out of 'b' digits before we're actually done, make
     sure the carries get propagated upward...  
   */
  used = USED(a);
  while(w && ix < used) {
    w += *pa;
    *pa++ = ACCUM(w);
    w = CARRYOUT(w);
    ++ix;
  }

  /* If there's an overall carry out, increase precision and include
     it.  We could have done this initially, but why touch the memory
     allocator unless we're sure we have to?
   */
  if(w) {
    if((res = s_mp_pad(a, used + 1)) != MP_OKAY)
      return res;

    DIGIT(a, ix) = w;  /* pa may not be valid after s_mp_pad() call */
  }

  return MP_OKAY;

} /* end s_mp_add() */

/* }}} */

/* {{{ s_mp_sub(a, b) */

/* Compute a = |a| - |b|, assumes |a| >= |b|                              */
mp_err   s_mp_sub(mp_int *a, mp_int *b)        /* magnitude subtract      */
{
  mp_word   w = 0;
  mp_digit *pa, *pb;
  mp_size   ix, used = USED(b);

  /*
    Subtract and propagate borrow.  Up to the precision of b, this
    accounts for the digits of b; after that, we just make sure the
    carries get to the right place.  This saves having to pad b out to
    the precision of a just to make the loops work right...
   */
  pa = DIGITS(a);
  pb = DIGITS(b);

  for(ix = 0; ix < used; ++ix) {
    w = (RADIX + *pa) - w - *pb++;
    *pa++ = ACCUM(w);
    w = CARRYOUT(w) ? 0 : 1;
  }

  used = USED(a);
  while(ix < used) {
    w = RADIX + *pa - w;
    *pa++ = ACCUM(w);
    w = CARRYOUT(w) ? 0 : 1;
    ++ix;
  }

  /* Clobber any leading zeroes we created    */
  s_mp_clamp(a);

  /* 
     If there was a borrow out, then |b| > |a| in violation
     of our input invariant.  We've already done the work,
     but we'll at least complain about it...
   */
  if(w)
    return MP_RANGE;
  else
    return MP_OKAY;

} /* end s_mp_sub() */

/* }}} */

mp_err   s_mp_reduce(mp_int *x, mp_int *m, mp_int *mu)
{
  mp_int   q;
  mp_err   res;
  mp_size  um = USED(m);

  if((res = mp_init_copy(&q, x)) != MP_OKAY)
    return res;

  s_mp_rshd(&q, um - 1);       /* q1 = x / b^(k-1)  */
  s_mp_mul(&q, mu);            /* q2 = q1 * mu      */
  s_mp_rshd(&q, um + 1);       /* q3 = q2 / b^(k+1) */

  /* x = x mod b^(k+1), quick (no division) */
  s_mp_mod_2d(x, (mp_digit)(DIGIT_BIT * (um + 1)));

  /* q = q * m mod b^(k+1), quick (no division), uses the short multiplier */
#ifndef SHRT_MUL
  s_mp_mul(&q, m);
  s_mp_mod_2d(&q, (mp_digit)(DIGIT_BIT * (um + 1)));
#else
  s_mp_mul_dig(&q, m, um + 1);
#endif  

  /* x = x - q */
  if((res = mp_sub(x, &q, x)) != MP_OKAY)
    goto CLEANUP;

  /* If x < 0, add b^(k+1) to it */
  if(mp_cmp_z(x) < 0) {
    mp_set(&q, 1);
    if((res = s_mp_lshd(&q, um + 1)) != MP_OKAY)
      goto CLEANUP;
    if((res = mp_add(x, &q, x)) != MP_OKAY)
      goto CLEANUP;
  }

  /* Back off if it's too big */
  while(mp_cmp(x, m) >= 0) {
    if((res = s_mp_sub(x, m)) != MP_OKAY)
      break;
  }

 CLEANUP:
  mp_clear(&q);

  return res;

} /* end s_mp_reduce() */



/* {{{ s_mp_mul(a, b) */

/* Compute a = |a| * |b|                                                  */
mp_err   s_mp_mul(mp_int *a, mp_int *b)
{
  mp_word   w, k = 0;
  mp_int    tmp;
  mp_err    res;
  mp_size   ix, jx, ua = USED(a), ub = USED(b);
  mp_digit *pa, *pb, *pt, *pbt;

  if((res = mp_init_size(&tmp, ua + ub)) != MP_OKAY)
    return res;

  /* This has the effect of left-padding with zeroes... */
  USED(&tmp) = ua + ub;

  /* We're going to need the base value each iteration */
  pbt = DIGITS(&tmp);

  /* Outer loop:  Digits of b */

  pb = DIGITS(b);
  for(ix = 0; ix < ub; ++ix, ++pb) {
    if(*pb == 0) 
      continue;

    /* Inner product:  Digits of a */
    pa = DIGITS(a);
    for(jx = 0; jx < ua; ++jx, ++pa) {
      pt = pbt + ix + jx;
      w = *pb * *pa + k + *pt;
      *pt = ACCUM(w);
      k = CARRYOUT(w);
    }

    pbt[ix + jx] = k;
    k = 0;
  }

  s_mp_clamp(&tmp);
  s_mp_exch(&tmp, a);

  mp_clear(&tmp);

  return MP_OKAY;

} /* end s_mp_mul() */

/* }}} */

/* {{{ s_mp_kmul(a, b, out, len) */

#if 0
void   s_mp_kmul(mp_digit *a, mp_digit *b, mp_digit *out, mp_size len)
{
  mp_word   w, k = 0;
  mp_size   ix, jx;
  mp_digit *pa, *pt;

  for(ix = 0; ix < len; ++ix, ++b) {
    if(*b == 0)
      continue;
    
    pa = a;
    for(jx = 0; jx < len; ++jx, ++pa) {
      pt = out + ix + jx;
      w = *b * *pa + k + *pt;
      *pt = ACCUM(w);
      k = CARRYOUT(w);
    }

    out[ix + jx] = k;
    k = 0;
  }

} /* end s_mp_kmul() */
#endif

/* }}} */

/* {{{ s_mp_sqr(a) */

/*
  Computes the square of a, in place.  This can be done more
  efficiently than a general multiplication, because many of the
  computation steps are redundant when squaring.  The inner product
  step is a bit more complicated, but we save a fair number of
  iterations of the multiplication loop.
 */
#if MP_SQUARE
mp_err   s_mp_sqr(mp_int *a)
{
  mp_word  w, k = 0;
  mp_int   tmp;
  mp_err   res;
  mp_size  ix, jx, kx, used = USED(a);
  mp_digit *pa1, *pa2, *pt, *pbt;

  if((res = mp_init_size(&tmp, 2 * used)) != MP_OKAY)
    return res;

  /* Left-pad with zeroes */
  USED(&tmp) = 2 * used;

  /* We need the base value each time through the loop */
  pbt = DIGITS(&tmp);

  pa1 = DIGITS(a);
  for(ix = 0; ix < used; ++ix, ++pa1) {
    if(*pa1 == 0)
      continue;

    w = DIGIT(&tmp, ix + ix) + (*pa1 * *pa1);

    pbt[ix + ix] = ACCUM(w);
    k = CARRYOUT(w);

    /*
      The inner product is computed as:

         (C, S) = t[i,j] + 2 a[i] a[j] + C

      This can overflow what can be represented in an mp_word, and
      since C arithmetic does not provide any way to check for
      overflow, we have to check explicitly for overflow conditions
      before they happen.
     */
    for(jx = ix + 1, pa2 = DIGITS(a) + jx; jx < used; ++jx, ++pa2) {
      mp_word  u = 0, v;
      
      /* Store this in a temporary to avoid indirections later */
      pt = pbt + ix + jx;

      /* Compute the multiplicative step */
      w = *pa1 * *pa2;

      /* If w is more than half MP_WORD_MAX, the doubling will
	 overflow, and we need to record a carry out into the next
	 word */
      u = (w >> (MP_WORD_BIT - 1)) & 1;

      /* Double what we've got, overflow will be ignored as defined
	 for C arithmetic (we've already noted if it is to occur)
       */
      w *= 2;

      /* Compute the additive step */
      v = *pt + k;

      /* If we do not already have an overflow carry, check to see
	 if the addition will cause one, and set the carry out if so 
       */
      u |= ((MP_WORD_MAX - v) < w);

      /* Add in the rest, again ignoring overflow */
      w += v;

      /* Set the i,j digit of the output */
      *pt = ACCUM(w);

      /* Save carry information for the next iteration of the loop.
	 This is why k must be an mp_word, instead of an mp_digit */
      k = CARRYOUT(w) | (u << DIGIT_BIT);

    } /* for(jx ...) */

    /* Set the last digit in the cycle and reset the carry */
    k = DIGIT(&tmp, ix + jx) + k;
    pbt[ix + jx] = ACCUM(k);
    k = CARRYOUT(k);

    /* If we are carrying out, propagate the carry to the next digit
       in the output.  This may cascade, so we have to be somewhat
       circumspect -- but we will have enough precision in the output
       that we won't overflow 
     */
    kx = 1;
    while(k) {
      k = pbt[ix + jx + kx] + 1;
      pbt[ix + jx + kx] = ACCUM(k);
      k = CARRYOUT(k);
      ++kx;
    }
  } /* for(ix ...) */

  s_mp_clamp(&tmp);
  s_mp_exch(&tmp, a);

  mp_clear(&tmp);

  return MP_OKAY;

} /* end s_mp_sqr() */
#endif

/* }}} */

/* {{{ s_mp_div(a, b) */

/*
  s_mp_div(a, b)

  Compute a = a / b and b = a mod b.  Assumes b > a.
 */

mp_err   s_mp_div(mp_int *a, mp_int *b)
{
  mp_int   quot, rem, t;
  mp_word  q;
  mp_err   res;
  mp_digit d;
  int      ix;

  if(mp_cmp_z(b) == 0)
    return MP_RANGE;

  /* Shortcut if b is power of two */
  if((ix = s_mp_ispow2(b)) >= 0) {
    mp_copy(a, b);  /* need this for remainder */
    s_mp_div_2d(a, (mp_digit)ix);
    s_mp_mod_2d(b, (mp_digit)ix);

    return MP_OKAY;
  }

  /* Allocate space to store the quotient */
  if((res = mp_init_size(&quot, USED(a))) != MP_OKAY)
    return res;

  /* A working temporary for division     */
  if((res = mp_init_size(&t, USED(a))) != MP_OKAY)
    goto T;

  /* Allocate space for the remainder     */
  if((res = mp_init_size(&rem, USED(a))) != MP_OKAY)
    goto REM;

  /* Normalize to optimize guessing       */
  d = s_mp_norm(a, b);

  /* Perform the division itself...woo!   */
  ix = USED(a) - 1;

  while(ix >= 0) {
    /* Find a partial substring of a which is at least b */
    while(s_mp_cmp(&rem, b) < 0 && ix >= 0) {
      if((res = s_mp_lshd(&rem, 1)) != MP_OKAY) 
	goto CLEANUP;

      if((res = s_mp_lshd(&quot, 1)) != MP_OKAY)
	goto CLEANUP;

      DIGIT(&rem, 0) = DIGIT(a, ix);
      s_mp_clamp(&rem);
      --ix;
    }

    /* If we didn't find one, we're finished dividing    */
    if(s_mp_cmp(&rem, b) < 0) 
      break;    

    /* Compute a guess for the next quotient digit       */
    q = DIGIT(&rem, USED(&rem) - 1);
    if(q <= DIGIT(b, USED(b) - 1) && USED(&rem) > 1)
      q = (q << DIGIT_BIT) | DIGIT(&rem, USED(&rem) - 2);

    q /= DIGIT(b, USED(b) - 1);

    /* The guess can be as much as RADIX + 1 */
    if(q >= RADIX)
      q = RADIX - 1;

    /* See what that multiplies out to                   */
    mp_copy(b, &t);
    if((res = s_mp_mul_d(&t, q)) != MP_OKAY)
      goto CLEANUP;

    /* 
       If it's too big, back it off.  We should not have to do this
       more than once, or, in rare cases, twice.  Knuth describes a
       method by which this could be reduced to a maximum of once, but
       I didn't implement that here.
     */
    while(s_mp_cmp(&t, &rem) > 0) {
      --q;
      s_mp_sub(&t, b);
    }

    /* At this point, q should be the right next digit   */
    if((res = s_mp_sub(&rem, &t)) != MP_OKAY)
      goto CLEANUP;

    /*
      Include the digit in the quotient.  We allocated enough memory
      for any quotient we could ever possibly get, so we should not
      have to check for failures here
     */
    DIGIT(&quot, 0) = q;
  }

  /* Denormalize remainder                */
  if(d != 0) 
    s_mp_div_2d(&rem, d);

  s_mp_clamp(&quot);
  s_mp_clamp(&rem);

  /* Copy quotient back to output         */
  s_mp_exch(&quot, a);
  
  /* Copy remainder back to output        */
  s_mp_exch(&rem, b);

CLEANUP:
  mp_clear(&rem);
REM:
  mp_clear(&t);
T:
  mp_clear(&quot);

  return res;

} /* end s_mp_div() */

/* }}} */

/* {{{ s_mp_2expt(a, k) */

mp_err   s_mp_2expt(mp_int *a, mp_digit k)
{
  mp_err    res;
  mp_size   dig, bit;

  dig = k / DIGIT_BIT;
  bit = k % DIGIT_BIT;

  mp_zero(a);
  if((res = s_mp_pad(a, dig + 1)) != MP_OKAY)
    return res;
  
  DIGIT(a, dig) |= (1 << bit);

  return MP_OKAY;

} /* end s_mp_2expt() */

/* }}} */


/* }}} */

/* }}} */

/* {{{ Primitive comparisons */

/* {{{ s_mp_cmp(a, b) */

/* Compare |a| <=> |b|, return 0 if equal, <0 if a<b, >0 if a>b           */
int      s_mp_cmp(mp_int *a, mp_int *b)
{
  mp_size   ua = USED(a), ub = USED(b);

  if(ua > ub)
    return MP_GT;
  else if(ua < ub)
    return MP_LT;
  else {
    int      ix = ua - 1;
    mp_digit *ap = DIGITS(a) + ix, *bp = DIGITS(b) + ix;

    while(ix >= 0) {
      if(*ap > *bp)
	return MP_GT;
      else if(*ap < *bp)
	return MP_LT;

      --ap; --bp; --ix;
    }

    return MP_EQ;
  }

} /* end s_mp_cmp() */

/* }}} */

/* {{{ s_mp_cmp_d(a, d) */

/* Compare |a| <=> d, return 0 if equal, <0 if a<d, >0 if a>d             */
int      s_mp_cmp_d(mp_int *a, mp_digit d)
{
  mp_size  ua = USED(a);
  mp_digit *ap = DIGITS(a);

  if(ua > 1)
    return MP_GT;

  if(*ap < d) 
    return MP_LT;
  else if(*ap > d)
    return MP_GT;
  else
    return MP_EQ;

} /* end s_mp_cmp_d() */

/* }}} */

/* {{{ s_mp_ispow2(v) */

/*
  Returns -1 if the value is not a power of two; otherwise, it returns
  k such that v = 2^k, i.e. lg(v).
 */
int      s_mp_ispow2(mp_int *v)
{
  mp_digit d, *dp;
  mp_size  uv = USED(v);
  int      extra = 0, ix;

  d = DIGIT(v, uv - 1); /* most significant digit of v */

  while(d && ((d & 1) == 0)) {
    d >>= 1;
    ++extra;
  }

  if(d == 1) {
    ix = uv - 2;
    dp = DIGITS(v) + ix;

    while(ix >= 0) {
      if(*dp)
	return -1; /* not a power of two */

      --dp; --ix;
    }

    return ((uv - 1) * DIGIT_BIT) + extra;
  } 

  return -1;

} /* end s_mp_ispow2() */

/* }}} */

/* {{{ s_mp_ispow2d(d) */

int      s_mp_ispow2d(mp_digit d)
{
  int   pow = 0;

  while((d & 1) == 0) {
    ++pow; d >>= 1;
  }

  if(d == 1)
    return pow;

  return -1;

} /* end s_mp_ispow2d() */

/* }}} */

/* }}} */

/* {{{ Primitive I/O helpers */

/* {{{ s_mp_tovalue(ch, r) */

/*
  Convert the given character to its digit value, in the given radix.
  If the given character is not understood in the given radix, -1 is
  returned.  Otherwise the digit's numeric value is returned.

  The results will be odd if you use a radix < 2 or > 62, you are
  expected to know what you're up to.
 */
int      s_mp_tovalue(char ch, int r)
{
  int    val, xch;
  
  if(r > 36)
    xch = ch;
  else
    xch = toupper(ch);

  if(isdigit(xch))
    val = xch - '0';
  else if(isupper(xch))
    val = xch - 'A' + 10;
  else if(islower(xch))
    val = xch - 'a' + 36;
  else if(xch == '+')
    val = 62;
  else if(xch == '/')
    val = 63;
  else 
    return -1;

  if(val < 0 || val >= r)
    return -1;

  return val;

} /* end s_mp_tovalue() */

/* }}} */

/* {{{ s_mp_todigit(val, r, low) */

/*
  Convert val to a radix-r digit, if possible.  If val is out of range
  for r, returns zero.  Otherwise, returns an ASCII character denoting
  the value in the given radix.

  The results may be odd if you use a radix < 2 or > 64, you are
  expected to know what you're doing.
 */
  
char     s_mp_todigit(int val, int r, int low)
{
  char   ch;

  if(val < 0 || val >= r)
    return 0;

  ch = s_dmap_1[val];

  if(r <= 36 && low)
    ch = tolower(ch);

  return ch;

} /* end s_mp_todigit() */

/* }}} */

/* {{{ s_mp_outlen(bits, radix) */

/* 
   Return an estimate for how long a string is needed to hold a radix
   r representation of a number with 'bits' significant bits.

   Does not include space for a sign or a NUL terminator.
 */
int      s_mp_outlen(int bits, int r)
{
  return (int)((double)bits * LOG_V_2(r));

} /* end s_mp_outlen() */

/* }}} */

/* }}} */

/*------------------------------------------------------------------------*/
/* HERE THERE BE DRAGONS                                                  */
/* crc==4242132123, version==2, Sat Feb 02 06:43:52 2002 */