|
Packit |
5c3484 |
/* mpn_dcpi1_divappr_q -- divide-and-conquer division, returning approximate
|
|
Packit |
5c3484 |
quotient. The quotient returned is either correct, or one too large.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
Contributed to the GNU project by Torbjorn Granlund.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
THE FUNCTIONS IN THIS FILE ARE INTERNAL WITH MUTABLE INTERFACES. IT IS ONLY
|
|
Packit |
5c3484 |
SAFE TO REACH THEM THROUGH DOCUMENTED INTERFACES. IN FACT, IT IS ALMOST
|
|
Packit |
5c3484 |
GUARANTEED THAT THEY WILL CHANGE OR DISAPPEAR IN A FUTURE GMP RELEASE.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
Copyright 2006, 2007, 2009, 2010 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 |
#include "longlong.h"
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
mp_limb_t
|
|
Packit |
5c3484 |
mpn_dcpi1_divappr_q_n (mp_ptr qp, mp_ptr np, mp_srcptr dp, mp_size_t n,
|
|
Packit |
5c3484 |
gmp_pi1_t *dinv, mp_ptr tp)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
mp_size_t lo, hi;
|
|
Packit |
5c3484 |
mp_limb_t cy, qh, ql;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
lo = n >> 1; /* floor(n/2) */
|
|
Packit |
5c3484 |
hi = n - lo; /* ceil(n/2) */
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
if (BELOW_THRESHOLD (hi, DC_DIV_QR_THRESHOLD))
|
|
Packit |
5c3484 |
qh = mpn_sbpi1_div_qr (qp + lo, np + 2 * lo, 2 * hi, dp + lo, hi, dinv->inv32);
|
|
Packit |
5c3484 |
else
|
|
Packit |
5c3484 |
qh = mpn_dcpi1_div_qr_n (qp + lo, np + 2 * lo, dp + lo, hi, dinv, tp);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
mpn_mul (tp, qp + lo, hi, dp, lo);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
cy = mpn_sub_n (np + lo, np + lo, tp, n);
|
|
Packit |
5c3484 |
if (qh != 0)
|
|
Packit |
5c3484 |
cy += mpn_sub_n (np + n, np + n, dp, lo);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
while (cy != 0)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
qh -= mpn_sub_1 (qp + lo, qp + lo, hi, 1);
|
|
Packit |
5c3484 |
cy -= mpn_add_n (np + lo, np + lo, dp, n);
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
if (BELOW_THRESHOLD (lo, DC_DIVAPPR_Q_THRESHOLD))
|
|
Packit |
5c3484 |
ql = mpn_sbpi1_divappr_q (qp, np + hi, 2 * lo, dp + hi, lo, dinv->inv32);
|
|
Packit |
5c3484 |
else
|
|
Packit |
5c3484 |
ql = mpn_dcpi1_divappr_q_n (qp, np + hi, dp + hi, lo, dinv, tp);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
if (UNLIKELY (ql != 0))
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
mp_size_t i;
|
|
Packit |
5c3484 |
for (i = 0; i < lo; i++)
|
|
Packit |
5c3484 |
qp[i] = GMP_NUMB_MASK;
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
return qh;
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
mp_limb_t
|
|
Packit |
5c3484 |
mpn_dcpi1_divappr_q (mp_ptr qp, mp_ptr np, mp_size_t nn,
|
|
Packit |
5c3484 |
mp_srcptr dp, mp_size_t dn, gmp_pi1_t *dinv)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
mp_size_t qn;
|
|
Packit |
5c3484 |
mp_limb_t qh, cy, qsave;
|
|
Packit |
5c3484 |
mp_ptr tp;
|
|
Packit |
5c3484 |
TMP_DECL;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
TMP_MARK;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASSERT (dn >= 6);
|
|
Packit |
5c3484 |
ASSERT (nn > dn);
|
|
Packit |
5c3484 |
ASSERT (dp[dn-1] & GMP_NUMB_HIGHBIT);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
qn = nn - dn;
|
|
Packit |
5c3484 |
qp += qn;
|
|
Packit |
5c3484 |
np += nn;
|
|
Packit |
5c3484 |
dp += dn;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
if (qn >= dn)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
qn++; /* pretend we'll need an extra limb */
|
|
Packit |
5c3484 |
/* Reduce qn mod dn without division, optimizing small operations. */
|
|
Packit |
5c3484 |
do
|
|
Packit |
5c3484 |
qn -= dn;
|
|
Packit |
5c3484 |
while (qn > dn);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
qp -= qn; /* point at low limb of next quotient block */
|
|
Packit |
5c3484 |
np -= qn; /* point in the middle of partial remainder */
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
tp = TMP_SALLOC_LIMBS (dn);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* Perform the typically smaller block first. */
|
|
Packit |
5c3484 |
if (qn == 1)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
mp_limb_t q, n2, n1, n0, d1, d0;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* Handle qh up front, for simplicity. */
|
|
Packit |
5c3484 |
qh = mpn_cmp (np - dn + 1, dp - dn, dn) >= 0;
|
|
Packit |
5c3484 |
if (qh)
|
|
Packit |
5c3484 |
ASSERT_NOCARRY (mpn_sub_n (np - dn + 1, np - dn + 1, dp - dn, dn));
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* A single iteration of schoolbook: One 3/2 division,
|
|
Packit |
5c3484 |
followed by the bignum update and adjustment. */
|
|
Packit |
5c3484 |
n2 = np[0];
|
|
Packit |
5c3484 |
n1 = np[-1];
|
|
Packit |
5c3484 |
n0 = np[-2];
|
|
Packit |
5c3484 |
d1 = dp[-1];
|
|
Packit |
5c3484 |
d0 = dp[-2];
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASSERT (n2 < d1 || (n2 == d1 && n1 <= d0));
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
if (UNLIKELY (n2 == d1) && n1 == d0)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
q = GMP_NUMB_MASK;
|
|
Packit |
5c3484 |
cy = mpn_submul_1 (np - dn, dp - dn, dn, q);
|
|
Packit |
5c3484 |
ASSERT (cy == n2);
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
else
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
udiv_qr_3by2 (q, n1, n0, n2, n1, n0, d1, d0, dinv->inv32);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
if (dn > 2)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
mp_limb_t cy, cy1;
|
|
Packit |
5c3484 |
cy = mpn_submul_1 (np - dn, dp - dn, dn - 2, q);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
cy1 = n0 < cy;
|
|
Packit |
5c3484 |
n0 = (n0 - cy) & GMP_NUMB_MASK;
|
|
Packit |
5c3484 |
cy = n1 < cy1;
|
|
Packit |
5c3484 |
n1 = (n1 - cy1) & GMP_NUMB_MASK;
|
|
Packit |
5c3484 |
np[-2] = n0;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
if (UNLIKELY (cy != 0))
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
n1 += d1 + mpn_add_n (np - dn, np - dn, dp - dn, dn - 1);
|
|
Packit |
5c3484 |
qh -= (q == 0);
|
|
Packit |
5c3484 |
q = (q - 1) & GMP_NUMB_MASK;
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
else
|
|
Packit |
5c3484 |
np[-2] = n0;
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
np[-1] = n1;
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
qp[0] = q;
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
else
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
if (qn == 2)
|
|
Packit |
5c3484 |
qh = mpn_divrem_2 (qp, 0L, np - 2, 4, dp - 2);
|
|
Packit |
5c3484 |
else if (BELOW_THRESHOLD (qn, DC_DIV_QR_THRESHOLD))
|
|
Packit |
5c3484 |
qh = mpn_sbpi1_div_qr (qp, np - qn, 2 * qn, dp - qn, qn, dinv->inv32);
|
|
Packit |
5c3484 |
else
|
|
Packit |
5c3484 |
qh = mpn_dcpi1_div_qr_n (qp, np - qn, dp - qn, qn, dinv, tp);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
if (qn != dn)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
if (qn > dn - qn)
|
|
Packit |
5c3484 |
mpn_mul (tp, qp, qn, dp - dn, dn - qn);
|
|
Packit |
5c3484 |
else
|
|
Packit |
5c3484 |
mpn_mul (tp, dp - dn, dn - qn, qp, qn);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
cy = mpn_sub_n (np - dn, np - dn, tp, dn);
|
|
Packit |
5c3484 |
if (qh != 0)
|
|
Packit |
5c3484 |
cy += mpn_sub_n (np - dn + qn, np - dn + qn, dp - dn, dn - qn);
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
while (cy != 0)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
qh -= mpn_sub_1 (qp, qp, qn, 1);
|
|
Packit |
5c3484 |
cy -= mpn_add_n (np - dn, np - dn, dp - dn, dn);
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
qn = nn - dn - qn + 1;
|
|
Packit |
5c3484 |
while (qn > dn)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
qp -= dn;
|
|
Packit |
5c3484 |
np -= dn;
|
|
Packit |
5c3484 |
mpn_dcpi1_div_qr_n (qp, np - dn, dp - dn, dn, dinv, tp);
|
|
Packit |
5c3484 |
qn -= dn;
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
/* Since we pretended we'd need an extra quotient limb before, we now
|
|
Packit |
5c3484 |
have made sure the code above left just dn-1=qn quotient limbs to
|
|
Packit |
5c3484 |
develop. Develop that plus a guard limb. */
|
|
Packit |
5c3484 |
qn--;
|
|
Packit |
5c3484 |
qp -= qn;
|
|
Packit |
5c3484 |
np -= dn;
|
|
Packit |
5c3484 |
qsave = qp[qn];
|
|
Packit |
5c3484 |
mpn_dcpi1_divappr_q_n (qp, np - dn, dp - dn, dn, dinv, tp);
|
|
Packit |
5c3484 |
MPN_COPY_INCR (qp, qp + 1, qn);
|
|
Packit |
5c3484 |
qp[qn] = qsave;
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
else /* (qn < dn) */
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
mp_ptr q2p;
|
|
Packit |
5c3484 |
#if 0 /* not possible since we demand nn > dn */
|
|
Packit |
5c3484 |
if (qn == 0)
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
qh = mpn_cmp (np - dn, dp - dn, dn) >= 0;
|
|
Packit |
5c3484 |
if (qh)
|
|
Packit |
5c3484 |
mpn_sub_n (np - dn, np - dn, dp - dn, dn);
|
|
Packit |
5c3484 |
TMP_FREE;
|
|
Packit |
5c3484 |
return qh;
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
#endif
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
qp -= qn; /* point at low limb of next quotient block */
|
|
Packit |
5c3484 |
np -= qn; /* point in the middle of partial remainder */
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
q2p = TMP_SALLOC_LIMBS (qn + 1);
|
|
Packit |
5c3484 |
/* Should we at all check DC_DIVAPPR_Q_THRESHOLD here, or reply on
|
|
Packit |
5c3484 |
callers not to be silly? */
|
|
Packit |
5c3484 |
if (BELOW_THRESHOLD (qn, DC_DIVAPPR_Q_THRESHOLD))
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
qh = mpn_sbpi1_divappr_q (q2p, np - qn - 2, 2 * (qn + 1),
|
|
Packit |
5c3484 |
dp - (qn + 1), qn + 1, dinv->inv32);
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
else
|
|
Packit |
5c3484 |
{
|
|
Packit |
5c3484 |
/* It is tempting to use qp for recursive scratch and put quotient in
|
|
Packit |
5c3484 |
tp, but the recursive scratch needs one limb too many. */
|
|
Packit |
5c3484 |
tp = TMP_SALLOC_LIMBS (qn + 1);
|
|
Packit |
5c3484 |
qh = mpn_dcpi1_divappr_q_n (q2p, np - qn - 2, dp - (qn + 1), qn + 1, dinv, tp);
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
MPN_COPY (qp, q2p + 1, qn);
|
|
Packit |
5c3484 |
}
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
TMP_FREE;
|
|
Packit |
5c3484 |
return qh;
|
|
Packit |
5c3484 |
}
|