2
|
1 /* LibTomMath, multiple-precision integer library -- Tom St Denis |
|
2 * |
|
3 * LibTomMath is a library that provides multiple-precision |
|
4 * integer arithmetic as well as number theoretic functionality. |
|
5 * |
|
6 * The library was designed directly after the MPI library by |
|
7 * Michael Fromberger but has been written from scratch with |
|
8 * additional optimizations in place. |
|
9 * |
|
10 * The library is free for all purposes without any express |
|
11 * guarantee it works. |
|
12 * |
|
13 * Tom St Denis, [email protected], http://math.libtomcrypt.org |
|
14 */ |
|
15 #include <tommath.h> |
|
16 |
|
17 /* this is a modified version of fast_s_mp_mul_digs that only produces |
|
18 * output digits *above* digs. See the comments for fast_s_mp_mul_digs |
|
19 * to see how it works. |
|
20 * |
|
21 * This is used in the Barrett reduction since for one of the multiplications |
|
22 * only the higher digits were needed. This essentially halves the work. |
|
23 * |
|
24 * Based on Algorithm 14.12 on pp.595 of HAC. |
|
25 */ |
|
26 int |
|
27 fast_s_mp_mul_high_digs (mp_int * a, mp_int * b, mp_int * c, int digs) |
|
28 { |
|
29 int oldused, newused, res, pa, pb, ix; |
|
30 mp_word W[MP_WARRAY]; |
|
31 |
|
32 /* calculate size of product and allocate more space if required */ |
|
33 newused = a->used + b->used + 1; |
|
34 if (c->alloc < newused) { |
|
35 if ((res = mp_grow (c, newused)) != MP_OKAY) { |
|
36 return res; |
|
37 } |
|
38 } |
|
39 |
|
40 /* like the other comba method we compute the columns first */ |
|
41 pa = a->used; |
|
42 pb = b->used; |
|
43 memset (W + digs, 0, (pa + pb + 1 - digs) * sizeof (mp_word)); |
|
44 for (ix = 0; ix < pa; ix++) { |
|
45 { |
|
46 register mp_digit tmpx, *tmpy; |
|
47 register int iy; |
|
48 register mp_word *_W; |
|
49 |
|
50 /* work todo, that is we only calculate digits that are at "digs" or above */ |
|
51 iy = digs - ix; |
|
52 |
|
53 /* copy of word on the left of A[ix] * B[iy] */ |
|
54 tmpx = a->dp[ix]; |
|
55 |
|
56 /* alias for right side */ |
|
57 tmpy = b->dp + iy; |
|
58 |
|
59 /* alias for the columns of output. Offset to be equal to or above the |
|
60 * smallest digit place requested |
|
61 */ |
|
62 _W = W + digs; |
|
63 |
|
64 /* skip cases below zero where ix > digs */ |
|
65 if (iy < 0) { |
|
66 iy = abs(iy); |
|
67 tmpy += iy; |
|
68 _W += iy; |
|
69 iy = 0; |
|
70 } |
|
71 |
|
72 /* compute column products for digits above the minimum */ |
|
73 for (; iy < pb; iy++) { |
|
74 *_W++ += ((mp_word) tmpx) * ((mp_word)*tmpy++); |
|
75 } |
|
76 } |
|
77 } |
|
78 |
|
79 /* setup dest */ |
|
80 oldused = c->used; |
|
81 c->used = newused; |
|
82 |
|
83 /* now convert the array W downto what we need |
|
84 * |
|
85 * See comments in bn_fast_s_mp_mul_digs.c |
|
86 */ |
|
87 for (ix = digs + 1; ix < newused; ix++) { |
|
88 W[ix] += (W[ix - 1] >> ((mp_word) DIGIT_BIT)); |
|
89 c->dp[ix - 1] = (mp_digit) (W[ix - 1] & ((mp_word) MP_MASK)); |
|
90 } |
|
91 c->dp[newused - 1] = (mp_digit) (W[newused - 1] & ((mp_word) MP_MASK)); |
|
92 |
|
93 for (; ix < oldused; ix++) { |
|
94 c->dp[ix] = 0; |
|
95 } |
|
96 mp_clamp (c); |
|
97 return MP_OKAY; |
|
98 } |