• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 #include <tommath.h>
2 #ifdef BN_FAST_MP_MONTGOMERY_REDUCE_C
3 /* LibTomMath, multiple-precision integer library -- Tom St Denis
4  *
5  * LibTomMath is a library that provides multiple-precision
6  * integer arithmetic as well as number theoretic functionality.
7  *
8  * The library was designed directly after the MPI library by
9  * Michael Fromberger but has been written from scratch with
10  * additional optimizations in place.
11  *
12  * The library is free for all purposes without any express
13  * guarantee it works.
14  *
15  * Tom St Denis, tomstdenis@gmail.com, http://math.libtomcrypt.com
16  */
17 
18 /* computes xR**-1 == x (mod N) via Montgomery Reduction
19  *
20  * This is an optimized implementation of montgomery_reduce
21  * which uses the comba method to quickly calculate the columns of the
22  * reduction.
23  *
24  * Based on Algorithm 14.32 on pp.601 of HAC.
25 */
fast_mp_montgomery_reduce(mp_int * x,mp_int * n,mp_digit rho)26 int fast_mp_montgomery_reduce (mp_int * x, mp_int * n, mp_digit rho)
27 {
28   int     ix, res, olduse;
29   mp_word W[MP_WARRAY];
30 
31   /* get old used count */
32   olduse = x->used;
33 
34   /* grow a as required */
35   if (x->alloc < n->used + 1) {
36     if ((res = mp_grow (x, n->used + 1)) != MP_OKAY) {
37       return res;
38     }
39   }
40 
41   /* first we have to get the digits of the input into
42    * an array of double precision words W[...]
43    */
44   {
45     register mp_word *_W;
46     register mp_digit *tmpx;
47 
48     /* alias for the W[] array */
49     _W   = W;
50 
51     /* alias for the digits of  x*/
52     tmpx = x->dp;
53 
54     /* copy the digits of a into W[0..a->used-1] */
55     for (ix = 0; ix < x->used; ix++) {
56       *_W++ = *tmpx++;
57     }
58 
59     /* zero the high words of W[a->used..m->used*2] */
60     for (; ix < n->used * 2 + 1; ix++) {
61       *_W++ = 0;
62     }
63   }
64 
65   /* now we proceed to zero successive digits
66    * from the least significant upwards
67    */
68   for (ix = 0; ix < n->used; ix++) {
69     /* mu = ai * m' mod b
70      *
71      * We avoid a double precision multiplication (which isn't required)
72      * by casting the value down to a mp_digit.  Note this requires
73      * that W[ix-1] have  the carry cleared (see after the inner loop)
74      */
75     register mp_digit mu;
76     mu = (mp_digit) (((W[ix] & MP_MASK) * rho) & MP_MASK);
77 
78     /* a = a + mu * m * b**i
79      *
80      * This is computed in place and on the fly.  The multiplication
81      * by b**i is handled by offseting which columns the results
82      * are added to.
83      *
84      * Note the comba method normally doesn't handle carries in the
85      * inner loop In this case we fix the carry from the previous
86      * column since the Montgomery reduction requires digits of the
87      * result (so far) [see above] to work.  This is
88      * handled by fixing up one carry after the inner loop.  The
89      * carry fixups are done in order so after these loops the
90      * first m->used words of W[] have the carries fixed
91      */
92     {
93       register int iy;
94       register mp_digit *tmpn;
95       register mp_word *_W;
96 
97       /* alias for the digits of the modulus */
98       tmpn = n->dp;
99 
100       /* Alias for the columns set by an offset of ix */
101       _W = W + ix;
102 
103       /* inner loop */
104       for (iy = 0; iy < n->used; iy++) {
105           *_W++ += ((mp_word)mu) * ((mp_word)*tmpn++);
106       }
107     }
108 
109     /* now fix carry for next digit, W[ix+1] */
110     W[ix + 1] += W[ix] >> ((mp_word) DIGIT_BIT);
111   }
112 
113   /* now we have to propagate the carries and
114    * shift the words downward [all those least
115    * significant digits we zeroed].
116    */
117   {
118     register mp_digit *tmpx;
119     register mp_word *_W, *_W1;
120 
121     /* nox fix rest of carries */
122 
123     /* alias for current word */
124     _W1 = W + ix;
125 
126     /* alias for next word, where the carry goes */
127     _W = W + ++ix;
128 
129     for (; ix <= n->used * 2 + 1; ix++) {
130       *_W++ += *_W1++ >> ((mp_word) DIGIT_BIT);
131     }
132 
133     /* copy out, A = A/b**n
134      *
135      * The result is A/b**n but instead of converting from an
136      * array of mp_word to mp_digit than calling mp_rshd
137      * we just copy them in the right order
138      */
139 
140     /* alias for destination word */
141     tmpx = x->dp;
142 
143     /* alias for shifted double precision result */
144     _W = W + n->used;
145 
146     for (ix = 0; ix < n->used + 1; ix++) {
147       *tmpx++ = (mp_digit)(*_W++ & ((mp_word) MP_MASK));
148     }
149 
150     /* zero oldused digits, if the input a was larger than
151      * m->used+1 we'll have to clear the digits
152      */
153     for (; ix < olduse; ix++) {
154       *tmpx++ = 0;
155     }
156   }
157 
158   /* set the max used and clamp */
159   x->used = n->used + 1;
160   mp_clamp (x);
161 
162   /* if A >= m then A = A - m */
163   if (mp_cmp_mag (x, n) != MP_LT) {
164     return s_mp_sub (x, n, x);
165   }
166   return MP_OKAY;
167 }
168 #endif
169 
170 /* $Source: /cvs/libtom/libtommath/bn_fast_mp_montgomery_reduce.c,v $ */
171 /* $Revision: 1.3 $ */
172 /* $Date: 2006/03/31 14:18:44 $ */
173