|
Packit |
5c3484 |
/* mpn_toom2_sqr -- Square {ap,an}.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
Contributed to the GNU project by Torbjorn Granlund.
|
|
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 2006-2010, 2012, 2014 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 |
|
|
Packit |
5c3484 |
#include "gmp.h"
|
|
Packit |
5c3484 |
#include "gmp-impl.h"
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* Evaluate in: -1, 0, +inf
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
<-s--><--n-->
|
|
Packit |
5c3484 |
____ ______
|
|
Packit |
5c3484 |
|_a1_|___a0_|
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
v0 = a0 ^2 # A(0)^2
|
|
Packit |
5c3484 |
vm1 = (a0- a1)^2 # A(-1)^2
|
|
Packit |
5c3484 |
vinf= a1 ^2 # A(inf)^2
|
|
Packit |
5c3484 |
*/
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
#if TUNE_PROGRAM_BUILD || WANT_FAT_BINARY
|
|
Packit |
5c3484 |
#define MAYBE_sqr_toom2 1
|
|
Packit |
5c3484 |
#else
|
|
Packit |
5c3484 |
#define MAYBE_sqr_toom2 \
|
|
Packit |
5c3484 |
(SQR_TOOM3_THRESHOLD >= 2 * SQR_TOOM2_THRESHOLD)
|
|
Packit |
5c3484 |
#endif
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
#define TOOM2_SQR_REC(p, a, n, ws) \
|
|
Packit |
5c3484 |
do { \
|
|
Packit |
5c3484 |
if (! MAYBE_sqr_toom2 \
|
|
Packit |
5c3484 |
|| BELOW_THRESHOLD (n, SQR_TOOM2_THRESHOLD)) \
|
|
Packit |
5c3484 |
mpn_sqr_basecase (p, a, n); \
|
|
Packit |
5c3484 |
else \
|
|
Packit |
5c3484 |
mpn_toom2_sqr (p, a, n, ws); \
|
|
Packit |
5c3484 |
} while (0)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
void
|
|
Packit |
5c3484 |
mpn_toom2_sqr (mp_ptr pp,
|
|
Packit |
5c3484 |
mp_srcptr ap, mp_size_t an,
|
|
Packit |
5c3484 |
mp_ptr scratch)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
const int __gmpn_cpuvec_initialized = 1;
|
|
Packit |
5c3484 |
mp_size_t n, s;
|
|
Packit |
5c3484 |
mp_limb_t cy, cy2;
|
|
Packit |
5c3484 |
mp_ptr asm1;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
#define a0 ap
|
|
Packit |
5c3484 |
#define a1 (ap + n)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
s = an >> 1;
|
|
Packit |
5c3484 |
n = an - s;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASSERT (0 < s && s <= n && s >= n - 1);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
asm1 = pp;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* Compute asm1. */
|
|
Packit |
5c3484 |
if (s == n)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
if (mpn_cmp (a0, a1, n) < 0)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
mpn_sub_n (asm1, a1, a0, n);
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
else
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
mpn_sub_n (asm1, a0, a1, n);
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
else /* n - s == 1 */
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
if (a0[s] == 0 && mpn_cmp (a0, a1, s) < 0)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
mpn_sub_n (asm1, a1, a0, s);
|
|
Packit |
5c3484 |
asm1[s] = 0;
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
else
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
asm1[s] = a0[s] - mpn_sub_n (asm1, a0, a1, s);
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
#define v0 pp /* 2n */
|
|
Packit |
5c3484 |
#define vinf (pp + 2 * n) /* s+s */
|
|
Packit |
5c3484 |
#define vm1 scratch /* 2n */
|
|
Packit |
5c3484 |
#define scratch_out scratch + 2 * n
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* vm1, 2n limbs */
|
|
Packit |
5c3484 |
TOOM2_SQR_REC (vm1, asm1, n, scratch_out);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* vinf, s+s limbs */
|
|
Packit |
5c3484 |
TOOM2_SQR_REC (vinf, a1, s, scratch_out);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* v0, 2n limbs */
|
|
Packit |
5c3484 |
TOOM2_SQR_REC (v0, ap, n, scratch_out);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* H(v0) + L(vinf) */
|
|
Packit |
5c3484 |
cy = mpn_add_n (pp + 2 * n, v0 + n, vinf, n);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* L(v0) + H(v0) */
|
|
Packit |
5c3484 |
cy2 = cy + mpn_add_n (pp + n, pp + 2 * n, v0, n);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* L(vinf) + H(vinf) */
|
|
Packit |
5c3484 |
cy += mpn_add (pp + 2 * n, pp + 2 * n, n, vinf + n, s + s - n);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
cy -= mpn_sub_n (pp + n, pp + n, vm1, 2 * n);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASSERT (cy + 1 <= 3);
|
|
Packit |
5c3484 |
ASSERT (cy2 <= 2);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
MPN_INCR_U (pp + 2 * n, s + s, cy2);
|
|
Packit |
5c3484 |
if (LIKELY (cy <= 2))
|
|
Packit |
5c3484 |
MPN_INCR_U (pp + 3 * n, s + s - n, cy);
|
|
Packit |
5c3484 |
else
|
|
Packit |
5c3484 |
MPN_DECR_U (pp + 3 * n, s + s - n, 1);
|
|
Packit |
5c3484 |
}
|