bn_mp_karatsuba_mul.c

#include <tommath.h>
#ifdef BN_MP_KARATSUBA_MUL_C
/* LibTomMath, multiple-precision integer library -- Tom St Denis
 *
 * LibTomMath is a library that provides multiple-precision
 * integer arithmetic as well as number theoretic functionality.
 *
 * The library was designed directly after the MPI library by
 * Michael Fromberger but has been written from scratch with
 * additional optimizations in place.
 *
 * The library is free for all purposes without any express
 * guarantee it works.
 *
 * Tom St Denis, tomstdenis@gmail.com, http://libtom.org
 */

/* c = |a| * |b| using Karatsuba Multiplication using
 * three half size multiplications
 *
 * Let B represent the radix [e.g. 2**DIGIT_BIT] and
 * let n represent half of the number of digits in
 * the min(a,b)
 *
 * a = a1 * B**n + a0
 * b = b1 * B**n + b0
 *
 * Then, a * b =>
   a1b1 * B**2n + ((a1 + a0)(b1 + b0) - (a0b0 + a1b1)) * B + a0b0
 *
 * Note that a1b1 and a0b0 are used twice and only need to be
 * computed once.  So in total three half size (half # of
 * digit) multiplications are performed, a0b0, a1b1 and
 * (a1+b1)(a0+b0)
 *
 * Note that a multiplication of half the digits requires
 * 1/4th the number of single precision multiplications so in
 * total after one call 25% of the single precision multiplications
 * are saved.  Note also that the call to mp_mul can end up back
 * in this function if the a0, a1, b0, or b1 are above the threshold.
 * This is known as divide-and-conquer and leads to the famous
 * O(N**lg(3)) or O(N**1.584) work which is asymptopically lower than
 * the standard O(N**2) that the baseline/comba methods use.
 * Generally though the overhead of this method doesn't pay off
 * until a certain size (N ~ 80) is reached.
 */
int mp_karatsuba_mul(mp_int *a, mp_int *b, mp_int *c)
{
   mp_int  x0, x1, y0, y1, t1, x0y0, x1y1;
   int     B, err;

#ifdef USE_OPEN_MP_NOT
   int e1,e2;
#endif
   /* default the return code to an error */
   err = MP_MEM;

   /* min # of digits */
   B = MIN(a->used, b->used);
   if(B < KARATSUBA_MUL_CUTOFF){
      return mp_mul(a,b,c);
   }
   /* now divide in two */
   B = B >> 1;

   /* init copy all the temps */
   if (mp_init_size(&x0, B) != MP_OKAY)
      goto ERR;
   if (mp_init_size(&x1, a->used - B) != MP_OKAY)
      goto X0;
   if (mp_init_size(&y0, B) != MP_OKAY)
      goto X1;
   if (mp_init_size(&y1, b->used - B) != MP_OKAY)
      goto Y0;

   /* init temps */
   if (mp_init_size(&t1, B * 2) != MP_OKAY)
      goto Y1;
   if (mp_init_size(&x0y0, B * 2) != MP_OKAY)
      goto T1;
   if (mp_init_size(&x1y1, B * 2) != MP_OKAY)
      goto X0Y0;

   /* now shift the digits */
   x0.used = y0.used = B;
   x1.used = a->used - B;
   y1.used = b->used - B;

   {
      register int x;
      register mp_digit *tmpa, *tmpb, *tmpx, *tmpy;

      /* we copy the digits directly instead of using higher level functions
       * since we also need to shift the digits
       */
      tmpa = a->dp;
      tmpb = b->dp;

      tmpx = x0.dp;
      tmpy = y0.dp;
      for (x = 0; x < B; x++) {
         *tmpx++ = *tmpa++;
         *tmpy++ = *tmpb++;
      }

      tmpx = x1.dp;
      for (x = B; x < a->used; x++) {
         *tmpx++ = *tmpa++;
      }

      tmpy = y1.dp;
      for (x = B; x < b->used; x++) {
         *tmpy++ = *tmpb++;
      }
   }

   /* only need to clamp the lower words since by definition the
    * upper words x1/y1 must have a known number of digits
    */
   mp_clamp(&x0);
   mp_clamp(&y0);

   /* now calc the products x0y0 and x1y1 */
   /* after this x0 is no longer required, free temp [x0==t2]! */
#ifdef USE_OPEN_MP_NOT
#include <omp.h>
   #pragma omp atomic
   n--;
   #pragma omp task shared(x0y0) if(n>=2)
   e1 = mp_karatsuba_mul(&x0, &y0, &x0y0,n);
   #pragma omp task shared(x1y1) if(n>=2)
   e2 = mp_karatsuba_mul(&x1, &y1, &x1y1,n);
   #pragma omp taskwait
   if ((e1 != MP_OKAY) || (e2 != MP_OKAY))
      goto X1Y1;
#else
   if (mp_mul(&x0, &y0, &x0y0) != MP_OKAY)
      goto X1Y1;          /* x0y0 = x0*y0 */
   if (mp_mul(&x1, &y1, &x1y1) != MP_OKAY)
      goto X1Y1;          /* x1y1 = x1*y1 */
#endif

   /* now calc x1+x0 and y1+y0 */
   if (s_mp_add(&x1, &x0, &t1) != MP_OKAY)
      goto X1Y1;          /* t1 = x1 - x0 */
   if (s_mp_add(&y1, &y0, &x0) != MP_OKAY)
      goto X1Y1;          /* t2 = y1 - y0 */
   if (mp_mul(&t1, &x0, &t1) != MP_OKAY)
      goto X1Y1;          /* t1 = (x1 + x0) * (y1 + y0) */

   /* add x0y0 */
   if (mp_add(&x0y0, &x1y1, &x0) != MP_OKAY)
      goto X1Y1;          /* t2 = x0y0 + x1y1 */
   if (s_mp_sub(&t1, &x0, &t1) != MP_OKAY)
      goto X1Y1;          /* t1 = (x1+x0)*(y1+y0) - (x1y1 + x0y0) */

   /* shift by B */
   if (mp_lshd(&t1, B) != MP_OKAY)
      goto X1Y1;          /* t1 = (x0y0 + x1y1 - (x1-x0)*(y1-y0))<<B */
   if (mp_lshd(&x1y1, B * 2) != MP_OKAY)
      goto X1Y1;          /* x1y1 = x1y1 << 2*B */

   if (mp_add(&x0y0, &t1, &t1) != MP_OKAY)
      goto X1Y1;          /* t1 = x0y0 + t1 */
   if (mp_add(&t1, &x1y1, c) != MP_OKAY)
      goto X1Y1;          /* t1 = x0y0 + t1 + x1y1 */

   /* Algorithm succeeded set the return code to MP_OKAY */
   err = MP_OKAY;

X1Y1:
   mp_clear(&x1y1);
X0Y0:
   mp_clear(&x0y0);
T1:
   mp_clear(&t1);
Y1:
   mp_clear(&y1);
Y0:
   mp_clear(&y0);
X1:
   mp_clear(&x1);
X0:
   mp_clear(&x0);
ERR:
   return err;
}
#endif

/* $Source$ */
/* $Revision$ */
/* $Date$ */