Viewing file:  divrem_1.c (7.19 KB) -rw-r--r--
Select action/file-type:
 (+) |  (+) |  (+) | Code (+) | Session (+) |  (+) | SDB (+) |  (+) |  (+) |  (+) |  (+) |  (+) |
/* UltraSparc 64 mpn_divrem_1 -- mpn by limb division.
Copyright 1991, 1993, 1994, 1996, 1998-2001, 2003 Free Software Foundation,
Inc.
This file is part of the GNU MP Library.
The GNU MP Library is free software; you can redistribute it and/or modify
it under the terms of either:
* the GNU Lesser General Public License as published by the Free
Software Foundation; either version 3 of the License, or (at your
option) any later version.
or
* the GNU General Public License as published by the Free Software
Foundation; either version 2 of the License, or (at your option) any
later version.
or both in parallel, as here.
The GNU MP Library is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
for more details.
You should have received copies of the GNU General Public License and the
GNU Lesser General Public License along with the GNU MP Library. If not,
see https://www.gnu.org/licenses/. */
#include "gmp-impl.h"
#include "longlong.h"
#include "mpn/sparc64/sparc64.h"
/* 64-bit divisor 32-bit divisor
cycles/limb cycles/limb
(approx) (approx)
integer fraction integer fraction
Ultrasparc 2i: 160 160 122 96
*/
/* 32-bit divisors are treated in special case code. This requires 4 mulx
per limb instead of 8 in the general case.
For big endian systems we need HALF_ENDIAN_ADJ included in the src[i]
addressing, to get the two halves of each limb read in the correct order.
This is kept in an adj variable. Doing that measures about 4 c/l faster
than just writing HALF_ENDIAN_ADJ(i) in the integer loop. The latter
shouldn't be 6 cycles worth of work, but perhaps it doesn't schedule well
(on gcc 3.2.1 at least). The fraction loop doesn't seem affected, but we
still use a variable since that ought to work out best. */
mp_limb_t
mpn_divrem_1 (mp_ptr qp_limbptr, mp_size_t xsize_limbs,
mp_srcptr ap_limbptr, mp_size_t size_limbs, mp_limb_t d_limb)
{
mp_size_t total_size_limbs;
mp_size_t i;
ASSERT (xsize_limbs >= 0);
ASSERT (size_limbs >= 0);
ASSERT (d_limb != 0);
/* FIXME: What's the correct overlap rule when xsize!=0? */
ASSERT (MPN_SAME_OR_SEPARATE_P (qp_limbptr + xsize_limbs,
ap_limbptr, size_limbs));
total_size_limbs = size_limbs + xsize_limbs;
if (UNLIKELY (total_size_limbs == 0))
return 0;
/* udivx is good for total_size==1, and no need to bother checking
limb<divisor, since if that's likely the caller should check */
if (UNLIKELY (total_size_limbs == 1))
{
mp_limb_t a, q;
a = (LIKELY (size_limbs != 0) ? ap_limbptr[0] : 0);
q = a / d_limb;
qp_limbptr[0] = q;
return a - q*d_limb;
}
if (d_limb <= CNST_LIMB(0xFFFFFFFF))
{
mp_size_t size, xsize, total_size, adj;
unsigned *qp, n1, n0, q, r, nshift, norm_rmask;
mp_limb_t dinv_limb;
const unsigned *ap;
int norm, norm_rshift;
size = 2 * size_limbs;
xsize = 2 * xsize_limbs;
total_size = size + xsize;
ap = (unsigned *) ap_limbptr;
qp = (unsigned *) qp_limbptr;
qp += xsize;
r = 0; /* initial remainder */
if (LIKELY (size != 0))
{
n1 = ap[size-1 + HALF_ENDIAN_ADJ(1)];
/* If the length of the source is uniformly distributed, then
there's a 50% chance of the high 32-bits being zero, which we
can skip. */
if (n1 == 0)
{
n1 = ap[size-2 + HALF_ENDIAN_ADJ(0)];
total_size--;
size--;
ASSERT (size > 0); /* because always even */
qp[size + HALF_ENDIAN_ADJ(1)] = 0;
}
/* Skip a division if high < divisor (high quotient 0). Testing
here before before normalizing will still skip as often as
possible. */
if (n1 < d_limb)
{
r = n1;
size--;
qp[size + HALF_ENDIAN_ADJ(size)] = 0;
total_size--;
if (total_size == 0)
return r;
}
}
count_leading_zeros_32 (norm, d_limb);
norm -= 32;
d_limb <<= norm;
r <<= norm;
norm_rshift = 32 - norm;
norm_rmask = (norm == 0 ? 0 : 0xFFFFFFFF);
invert_half_limb (dinv_limb, d_limb);
if (LIKELY (size != 0))
{
i = size - 1;
adj = HALF_ENDIAN_ADJ (i);
n1 = ap[i + adj];
adj = -adj;
r |= ((n1 >> norm_rshift) & norm_rmask);
for ( ; i > 0; i--)
{
n0 = ap[i-1 + adj];
adj = -adj;
nshift = (n1 << norm) | ((n0 >> norm_rshift) & norm_rmask);
udiv_qrnnd_half_preinv (q, r, r, nshift, d_limb, dinv_limb);
qp[i + adj] = q;
n1 = n0;
}
nshift = n1 << norm;
udiv_qrnnd_half_preinv (q, r, r, nshift, d_limb, dinv_limb);
qp[0 + HALF_ENDIAN_ADJ(0)] = q;
}
qp -= xsize;
adj = HALF_ENDIAN_ADJ (0);
for (i = xsize-1; i >= 0; i--)
{
udiv_qrnnd_half_preinv (q, r, r, 0, d_limb, dinv_limb);
adj = -adj;
qp[i + adj] = q;
}
return r >> norm;
}
else
{
mp_srcptr ap;
mp_ptr qp;
mp_size_t size, xsize, total_size;
mp_limb_t d, n1, n0, q, r, dinv, nshift, norm_rmask;
int norm, norm_rshift;
ap = ap_limbptr;
qp = qp_limbptr;
size = size_limbs;
xsize = xsize_limbs;
total_size = total_size_limbs;
d = d_limb;
qp += total_size; /* above high limb */
r = 0; /* initial remainder */
if (LIKELY (size != 0))
{
/* Skip a division if high < divisor (high quotient 0). Testing
here before before normalizing will still skip as often as
possible. */
n1 = ap[size-1];
if (n1 < d)
{
r = n1;
*--qp = 0;
total_size--;
if (total_size == 0)
return r;
size--;
}
}
count_leading_zeros (norm, d);
d <<= norm;
r <<= norm;
norm_rshift = GMP_LIMB_BITS - norm;
norm_rmask = (norm == 0 ? 0 : ~CNST_LIMB(0));
invert_limb (dinv, d);
if (LIKELY (size != 0))
{
n1 = ap[size-1];
r |= ((n1 >> norm_rshift) & norm_rmask);
for (i = size-2; i >= 0; i--)
{
n0 = ap[i];
nshift = (n1 << norm) | ((n0 >> norm_rshift) & norm_rmask);
udiv_qrnnd_preinv (q, r, r, nshift, d, dinv);
*--qp = q;
n1 = n0;
}
nshift = n1 << norm;
udiv_qrnnd_preinv (q, r, r, nshift, d, dinv);
*--qp = q;
}
for (i = 0; i < xsize; i++)
{
udiv_qrnnd_preinv (q, r, r, CNST_LIMB(0), d, dinv);
*--qp = q;
}
return r >> norm;
}
}
|
