Line data    Source code

       1             : #include "tommath_private.h"
       2             : #ifdef BN_S_MP_MONTGOMERY_REDUCE_FAST_C
       3             : /* LibTomMath, multiple-precision integer library -- Tom St Denis */
       4             : /* SPDX-License-Identifier: Unlicense */
       5             : 
       6             : /* computes xR**-1 == x (mod N) via Montgomery Reduction
       7             :  *
       8             :  * This is an optimized implementation of montgomery_reduce
       9             :  * which uses the comba method to quickly calculate the columns of the
      10             :  * reduction.
      11             :  *
      12             :  * Based on Algorithm 14.32 on pp.601 of HAC.
      13             : */
      14      827134 : mp_err s_mp_montgomery_reduce_fast(mp_int *x, const mp_int *n, mp_digit rho)
      15             : {
      16       38380 :    int     ix, olduse;
      17       38380 :    mp_err  err;
      18       38380 :    mp_word W[MP_WARRAY];
      19             : 
      20      827134 :    if (x->used > MP_WARRAY) {
      21           0 :       return MP_VAL;
      22             :    }
      23             : 
      24             :    /* get old used count */
      25      827134 :    olduse = x->used;
      26             : 
      27             :    /* grow a as required */
      28      827134 :    if (x->alloc < (n->used + 1)) {
      29           0 :       if ((err = mp_grow(x, n->used + 1)) != MP_OKAY) {
      30           0 :          return err;
      31             :       }
      32             :    }
      33             : 
      34             :    /* first we have to get the digits of the input into
      35             :     * an array of double precision words W[...]
      36             :     */
      37             :    {
      38       38380 :       mp_word *_W;
      39       38380 :       mp_digit *tmpx;
      40             : 
      41             :       /* alias for the W[] array */
      42      827134 :       _W   = W;
      43             : 
      44             :       /* alias for the digits of  x*/
      45      827134 :       tmpx = x->dp;
      46             : 
      47             :       /* copy the digits of a into W[0..a->used-1] */
      48    48904913 :       for (ix = 0; ix < x->used; ix++) {
      49    48077779 :          *_W++ = *tmpx++;
      50             :       }
      51             : 
      52             :       /* zero the high words of W[a->used..m->used*2] */
      53      827134 :       if (ix < ((n->used * 2) + 1)) {
      54    28432462 :          MP_ZERO_BUFFER(_W, sizeof(mp_word) * (size_t)(((n->used * 2) + 1) - ix));
      55             :       }
      56             :    }
      57             : 
      58             :    /* now we proceed to zero successive digits
      59             :     * from the least significant upwards
      60             :     */
      61    25315123 :    for (ix = 0; ix < n->used; ix++) {
      62             :       /* mu = ai * m' mod b
      63             :        *
      64             :        * We avoid a double precision multiplication (which isn't required)
      65             :        * by casting the value down to a mp_digit.  Note this requires
      66             :        * that W[ix-1] have  the carry cleared (see after the inner loop)
      67             :        */
      68     1356900 :       mp_digit mu;
      69    24487989 :       mu = ((W[ix] & MP_MASK) * rho) & MP_MASK;
      70             : 
      71             :       /* a = a + mu * m * b**i
      72             :        *
      73             :        * This is computed in place and on the fly.  The multiplication
      74             :        * by b**i is handled by offseting which columns the results
      75             :        * are added to.
      76             :        *
      77             :        * Note the comba method normally doesn't handle carries in the
      78             :        * inner loop In this case we fix the carry from the previous
      79             :        * column since the Montgomery reduction requires digits of the
      80             :        * result (so far) [see above] to work.  This is
      81             :        * handled by fixing up one carry after the inner loop.  The
      82             :        * carry fixups are done in order so after these loops the
      83             :        * first m->used words of W[] have the carries fixed
      84             :        */
      85             :       {
      86     1356900 :          int iy;
      87     1356900 :          mp_digit *tmpn;
      88     1356900 :          mp_word *_W;
      89             : 
      90             :          /* alias for the digits of the modulus */
      91    24487989 :          tmpn = n->dp;
      92             : 
      93             :          /* Alias for the columns set by an offset of ix */
      94    24487989 :          _W = W + ix;
      95             : 
      96             :          /* inner loop */
      97   854274036 :          for (iy = 0; iy < n->used; iy++) {
      98   829786047 :             *_W++ += (mp_word)mu * (mp_word)*tmpn++;
      99             :          }
     100             :       }
     101             : 
     102             :       /* now fix carry for next digit, W[ix+1] */
     103    24487989 :       W[ix + 1] += W[ix] >> (mp_word)MP_DIGIT_BIT;
     104             :    }
     105             : 
     106             :    /* now we have to propagate the carries and
     107             :     * shift the words downward [all those least
     108             :     * significant digits we zeroed].
     109             :     */
     110             :    {
     111       38380 :       mp_digit *tmpx;
     112       38380 :       mp_word *_W, *_W1;
     113             : 
     114             :       /* nox fix rest of carries */
     115             : 
     116             :       /* alias for current word */
     117      827134 :       _W1 = W + ix;
     118             : 
     119             :       /* alias for next word, where the carry goes */
     120      827134 :       _W = W + ++ix;
     121             : 
     122    25315123 :       for (; ix < ((n->used * 2) + 1); ix++) {
     123    24487989 :          *_W++ += *_W1++ >> (mp_word)MP_DIGIT_BIT;
     124             :       }
     125             : 
     126             :       /* copy out, A = A/b**n
     127             :        *
     128             :        * The result is A/b**n but instead of converting from an
     129             :        * array of mp_word to mp_digit than calling mp_rshd
     130             :        * we just copy them in the right order
     131             :        */
     132             : 
     133             :       /* alias for destination word */
     134      827134 :       tmpx = x->dp;
     135             : 
     136             :       /* alias for shifted double precision result */
     137      827134 :       _W = W + n->used;
     138             : 
     139    26142257 :       for (ix = 0; ix < (n->used + 1); ix++) {
     140    25315123 :          *tmpx++ = *_W++ & (mp_word)MP_MASK;
     141             :       }
     142             : 
     143             :       /* zero oldused digits, if the input a was larger than
     144             :        * m->used+1 we'll have to clear the digits
     145             :        */
     146    23590735 :       MP_ZERO_DIGITS(tmpx, olduse - ix);
     147             :    }
     148             : 
     149             :    /* set the max used and clamp */
     150      827134 :    x->used = n->used + 1;
     151      827134 :    mp_clamp(x);
     152             : 
     153             :    /* if A >= m then A = A - m */
     154      827134 :    if (mp_cmp_mag(x, n) != MP_LT) {
     155           0 :       return s_mp_sub(x, n, x);
     156             :    }
     157      788754 :    return MP_OKAY;
     158             : }
     159             : #endif