comparison libtommath/bn_s_mp_karatsuba_sqr.c @ 1692:1051e4eea25a

Update LibTomMath to 1.2.0 (#84) * update C files * update other files * update headers * update makefiles * remove mp_set/get_double() * use ltm 1.2.0 API * update ltm_desc * use bundled tommath if system-tommath is too old * XMALLOC etc. were changed to MP_MALLOC etc.
author Steffen Jaeckel <s@jaeckel.eu>
date Tue, 26 May 2020 17:36:47 +0200
parents
children
comparison
equal deleted inserted replaced
1691:2d3745d58843 1692:1051e4eea25a
1 #include "tommath_private.h"
2 #ifdef BN_S_MP_KARATSUBA_SQR_C
3 /* LibTomMath, multiple-precision integer library -- Tom St Denis */
4 /* SPDX-License-Identifier: Unlicense */
5
6 /* Karatsuba squaring, computes b = a*a using three
7 * half size squarings
8 *
9 * See comments of karatsuba_mul for details. It
10 * is essentially the same algorithm but merely
11 * tuned to perform recursive squarings.
12 */
13 mp_err s_mp_karatsuba_sqr(const mp_int *a, mp_int *b)
14 {
15 mp_int x0, x1, t1, t2, x0x0, x1x1;
16 int B;
17 mp_err err = MP_MEM;
18
19 /* min # of digits */
20 B = a->used;
21
22 /* now divide in two */
23 B = B >> 1;
24
25 /* init copy all the temps */
26 if (mp_init_size(&x0, B) != MP_OKAY)
27 goto LBL_ERR;
28 if (mp_init_size(&x1, a->used - B) != MP_OKAY)
29 goto X0;
30
31 /* init temps */
32 if (mp_init_size(&t1, a->used * 2) != MP_OKAY)
33 goto X1;
34 if (mp_init_size(&t2, a->used * 2) != MP_OKAY)
35 goto T1;
36 if (mp_init_size(&x0x0, B * 2) != MP_OKAY)
37 goto T2;
38 if (mp_init_size(&x1x1, (a->used - B) * 2) != MP_OKAY)
39 goto X0X0;
40
41 {
42 int x;
43 mp_digit *dst, *src;
44
45 src = a->dp;
46
47 /* now shift the digits */
48 dst = x0.dp;
49 for (x = 0; x < B; x++) {
50 *dst++ = *src++;
51 }
52
53 dst = x1.dp;
54 for (x = B; x < a->used; x++) {
55 *dst++ = *src++;
56 }
57 }
58
59 x0.used = B;
60 x1.used = a->used - B;
61
62 mp_clamp(&x0);
63
64 /* now calc the products x0*x0 and x1*x1 */
65 if (mp_sqr(&x0, &x0x0) != MP_OKAY)
66 goto X1X1; /* x0x0 = x0*x0 */
67 if (mp_sqr(&x1, &x1x1) != MP_OKAY)
68 goto X1X1; /* x1x1 = x1*x1 */
69
70 /* now calc (x1+x0)**2 */
71 if (s_mp_add(&x1, &x0, &t1) != MP_OKAY)
72 goto X1X1; /* t1 = x1 - x0 */
73 if (mp_sqr(&t1, &t1) != MP_OKAY)
74 goto X1X1; /* t1 = (x1 - x0) * (x1 - x0) */
75
76 /* add x0y0 */
77 if (s_mp_add(&x0x0, &x1x1, &t2) != MP_OKAY)
78 goto X1X1; /* t2 = x0x0 + x1x1 */
79 if (s_mp_sub(&t1, &t2, &t1) != MP_OKAY)
80 goto X1X1; /* t1 = (x1+x0)**2 - (x0x0 + x1x1) */
81
82 /* shift by B */
83 if (mp_lshd(&t1, B) != MP_OKAY)
84 goto X1X1; /* t1 = (x0x0 + x1x1 - (x1-x0)*(x1-x0))<<B */
85 if (mp_lshd(&x1x1, B * 2) != MP_OKAY)
86 goto X1X1; /* x1x1 = x1x1 << 2*B */
87
88 if (mp_add(&x0x0, &t1, &t1) != MP_OKAY)
89 goto X1X1; /* t1 = x0x0 + t1 */
90 if (mp_add(&t1, &x1x1, b) != MP_OKAY)
91 goto X1X1; /* t1 = x0x0 + t1 + x1x1 */
92
93 err = MP_OKAY;
94
95 X1X1:
96 mp_clear(&x1x1);
97 X0X0:
98 mp_clear(&x0x0);
99 T2:
100 mp_clear(&t2);
101 T1:
102 mp_clear(&t1);
103 X1:
104 mp_clear(&x1);
105 X0:
106 mp_clear(&x0);
107 LBL_ERR:
108 return err;
109 }
110 #endif