|
Packit |
5c3484 |
/* Implementation of the algorithm for Toom-Cook 4.5-way.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
Contributed to the GNU project by Marco Bodrato.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
THE FUNCTION IN THIS FILE IS INTERNAL WITH A MUTABLE INTERFACE. IT IS ONLY
|
|
Packit |
5c3484 |
SAFE TO REACH IT THROUGH DOCUMENTED INTERFACES. IN FACT, IT IS ALMOST
|
|
Packit |
5c3484 |
GUARANTEED THAT IT WILL CHANGE OR DISAPPEAR IN A FUTURE GNU MP RELEASE.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
Copyright 2009, 2012 Free Software Foundation, Inc.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
This file is part of the GNU MP Library.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
The GNU MP Library is free software; you can redistribute it and/or modify
|
|
Packit |
5c3484 |
it under the terms of either:
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
* the GNU Lesser General Public License as published by the Free
|
|
Packit |
5c3484 |
Software Foundation; either version 3 of the License, or (at your
|
|
Packit |
5c3484 |
option) any later version.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
or
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
* the GNU General Public License as published by the Free Software
|
|
Packit |
5c3484 |
Foundation; either version 2 of the License, or (at your option) any
|
|
Packit |
5c3484 |
later version.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
or both in parallel, as here.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
The GNU MP Library is distributed in the hope that it will be useful, but
|
|
Packit |
5c3484 |
WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
Packit |
5c3484 |
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
|
Packit |
5c3484 |
for more details.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
You should have received copies of the GNU General Public License and the
|
|
Packit |
5c3484 |
GNU Lesser General Public License along with the GNU MP Library. If not,
|
|
Packit |
5c3484 |
see https://www.gnu.org/licenses/. */
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
#include "gmp.h"
|
|
Packit |
5c3484 |
#include "gmp-impl.h"
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* Toom-4.5, the splitting 5x4 unbalanced version.
|
|
Packit |
5c3484 |
Evaluate in: infinity, +4, -4, +2, -2, +1, -1, 0.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
<--s-><--n--><--n--><--n--><--n-->
|
|
Packit |
5c3484 |
____ ______ ______ ______ ______
|
|
Packit |
5c3484 |
|_a4_|__a3__|__a2__|__a1__|__a0__|
|
|
Packit |
5c3484 |
|b3_|__b2__|__b1__|__b0__|
|
|
Packit |
5c3484 |
<-t-><--n--><--n--><--n-->
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
*/
|
|
Packit |
5c3484 |
#define TOOM_54_MUL_N_REC(p, a, b, n, ws) \
|
|
Packit |
5c3484 |
do { mpn_mul_n (p, a, b, n); \
|
|
Packit |
5c3484 |
} while (0)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
#define TOOM_54_MUL_REC(p, a, na, b, nb, ws) \
|
|
Packit |
5c3484 |
do { mpn_mul (p, a, na, b, nb); \
|
|
Packit |
5c3484 |
} while (0)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
void
|
|
Packit |
5c3484 |
mpn_toom54_mul (mp_ptr pp,
|
|
Packit |
5c3484 |
mp_srcptr ap, mp_size_t an,
|
|
Packit |
5c3484 |
mp_srcptr bp, mp_size_t bn, mp_ptr scratch)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
mp_size_t n, s, t;
|
|
Packit |
5c3484 |
int sign;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/***************************** decomposition *******************************/
|
|
Packit |
5c3484 |
#define a4 (ap + 4 * n)
|
|
Packit |
5c3484 |
#define b3 (bp + 3 * n)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASSERT (an >= bn);
|
|
Packit |
5c3484 |
n = 1 + (4 * an >= 5 * bn ? (an - 1) / (size_t) 5 : (bn - 1) / (size_t) 4);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
s = an - 4 * n;
|
|
Packit |
5c3484 |
t = bn - 3 * n;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASSERT (0 < s && s <= n);
|
|
Packit |
5c3484 |
ASSERT (0 < t && t <= n);
|
|
Packit |
5c3484 |
/* Required by mpn_toom_interpolate_8pts. */
|
|
Packit |
5c3484 |
ASSERT ( s + t >= n );
|
|
Packit |
5c3484 |
ASSERT ( s + t > 4);
|
|
Packit |
5c3484 |
ASSERT ( n > 2);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
#define r8 pp /* 2n */
|
|
Packit |
5c3484 |
#define r7 scratch /* 3n+1 */
|
|
Packit |
5c3484 |
#define r5 (pp + 3*n) /* 3n+1 */
|
|
Packit |
5c3484 |
#define v0 (pp + 3*n) /* n+1 */
|
|
Packit |
5c3484 |
#define v1 (pp + 4*n+1) /* n+1 */
|
|
Packit |
5c3484 |
#define v2 (pp + 5*n+2) /* n+1 */
|
|
Packit |
5c3484 |
#define v3 (pp + 6*n+3) /* n+1 */
|
|
Packit |
5c3484 |
#define r3 (scratch + 3 * n + 1) /* 3n+1 */
|
|
Packit |
5c3484 |
#define r1 (pp + 7*n) /* s+t <= 2*n */
|
|
Packit |
5c3484 |
#define ws (scratch + 6 * n + 2) /* ??? */
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* Alloc also 3n+1 limbs for ws... mpn_toom_interpolate_8pts may
|
|
Packit |
5c3484 |
need all of them, when DO_mpn_sublsh_n usea a scratch */
|
|
Packit |
5c3484 |
/********************** evaluation and recursive calls *********************/
|
|
Packit |
5c3484 |
/* $\pm4$ */
|
|
Packit |
5c3484 |
sign = mpn_toom_eval_pm2exp (v2, v0, 4, ap, n, s, 2, pp)
|
|
Packit |
5c3484 |
^ mpn_toom_eval_pm2exp (v3, v1, 3, bp, n, t, 2, pp);
|
|
Packit |
5c3484 |
TOOM_54_MUL_N_REC(pp, v0, v1, n + 1, ws); /* A(-4)*B(-4) */
|
|
Packit |
5c3484 |
TOOM_54_MUL_N_REC(r3, v2, v3, n + 1, ws); /* A(+4)*B(+4) */
|
|
Packit |
5c3484 |
mpn_toom_couple_handling (r3, 2*n+1, pp, sign, n, 2, 4);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* $\pm1$ */
|
|
Packit |
5c3484 |
sign = mpn_toom_eval_pm1 (v2, v0, 4, ap, n, s, pp)
|
|
Packit |
5c3484 |
^ mpn_toom_eval_dgr3_pm1 (v3, v1, bp, n, t, pp);
|
|
Packit |
5c3484 |
TOOM_54_MUL_N_REC(pp, v0, v1, n + 1, ws); /* A(-1)*B(-1) */
|
|
Packit |
5c3484 |
TOOM_54_MUL_N_REC(r7, v2, v3, n + 1, ws); /* A(1)*B(1) */
|
|
Packit |
5c3484 |
mpn_toom_couple_handling (r7, 2*n+1, pp, sign, n, 0, 0);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* $\pm2$ */
|
|
Packit |
5c3484 |
sign = mpn_toom_eval_pm2 (v2, v0, 4, ap, n, s, pp)
|
|
Packit |
5c3484 |
^ mpn_toom_eval_dgr3_pm2 (v3, v1, bp, n, t, pp);
|
|
Packit |
5c3484 |
TOOM_54_MUL_N_REC(pp, v0, v1, n + 1, ws); /* A(-2)*B(-2) */
|
|
Packit |
5c3484 |
TOOM_54_MUL_N_REC(r5, v2, v3, n + 1, ws); /* A(+2)*B(+2) */
|
|
Packit |
5c3484 |
mpn_toom_couple_handling (r5, 2*n+1, pp, sign, n, 1, 2);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* A(0)*B(0) */
|
|
Packit |
5c3484 |
TOOM_54_MUL_N_REC(pp, ap, bp, n, ws);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* Infinity */
|
|
Packit |
5c3484 |
if (s > t) {
|
|
Packit |
5c3484 |
TOOM_54_MUL_REC(r1, a4, s, b3, t, ws);
|
|
Packit |
5c3484 |
} else {
|
|
Packit |
5c3484 |
TOOM_54_MUL_REC(r1, b3, t, a4, s, ws);
|
|
Packit |
5c3484 |
};
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
mpn_toom_interpolate_8pts (pp, n, r3, r7, s + t, ws);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
#undef a4
|
|
Packit |
5c3484 |
#undef b3
|
|
Packit |
5c3484 |
#undef r1
|
|
Packit |
5c3484 |
#undef r3
|
|
Packit |
5c3484 |
#undef r5
|
|
Packit |
5c3484 |
#undef v0
|
|
Packit |
5c3484 |
#undef v1
|
|
Packit |
5c3484 |
#undef v2
|
|
Packit |
5c3484 |
#undef v3
|
|
Packit |
5c3484 |
#undef r7
|
|
Packit |
5c3484 |
#undef r8
|
|
Packit |
5c3484 |
#undef ws
|
|
Packit |
5c3484 |
}
|