libtommath/bn_mp_tc_and.c
2018-12-27 09:44:29 +01:00

92 lines
2.0 KiB
C

#include "tommath_private.h"
#ifdef BN_MP_TC_AND_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.
*/
/* two complement and */
int mp_tc_and(const mp_int *a, const mp_int *b, mp_int *c)
{
int res = MP_OKAY, bits, abits, bbits;
int as = mp_isneg(a), bs = mp_isneg(b);
mp_int *mx = NULL, _mx, acpy, bcpy;
if ((as != MP_NO) || (bs != MP_NO)) {
abits = mp_count_bits(a);
bbits = mp_count_bits(b);
bits = MAX(abits, bbits);
res = mp_init_set_int(&_mx, 1uL);
if (res != MP_OKAY) {
goto end;
}
mx = &_mx;
res = mp_mul_2d(mx, bits + 1, mx);
if (res != MP_OKAY) {
goto end;
}
if (as != MP_NO) {
res = mp_init(&acpy);
if (res != MP_OKAY) {
goto end;
}
res = mp_add(mx, a, &acpy);
if (res != MP_OKAY) {
mp_clear(&acpy);
goto end;
}
a = &acpy;
}
if (bs != MP_NO) {
res = mp_init(&bcpy);
if (res != MP_OKAY) {
goto end;
}
res = mp_add(mx, b, &bcpy);
if (res != MP_OKAY) {
mp_clear(&bcpy);
goto end;
}
b = &bcpy;
}
}
res = mp_and(a, b, c);
if ((as != MP_NO) && (bs != MP_NO) && (res == MP_OKAY)) {
res = mp_sub(c, mx, c);
}
end:
if (a == &acpy) {
mp_clear(&acpy);
}
if (b == &bcpy) {
mp_clear(&bcpy);
}
if (mx == &_mx) {
mp_clear(mx);
}
return res;
}
#endif
/* ref: $Format:%D$ */
/* git commit: $Format:%H$ */
/* commit time: $Format:%ai$ */