186d7f5d3SJohn Marino /* mpn_mod_34lsub1 -- remainder modulo 2^(GMP_NUMB_BITS*3/4)-1.
286d7f5d3SJohn Marino
386d7f5d3SJohn Marino THE FUNCTIONS IN THIS FILE ARE FOR INTERNAL USE ONLY. THEY'RE ALMOST
486d7f5d3SJohn Marino CERTAIN TO BE SUBJECT TO INCOMPATIBLE CHANGES OR DISAPPEAR COMPLETELY IN
586d7f5d3SJohn Marino FUTURE GNU MP RELEASES.
686d7f5d3SJohn Marino
786d7f5d3SJohn Marino Copyright 2000, 2001, 2002 Free Software Foundation, Inc.
886d7f5d3SJohn Marino
986d7f5d3SJohn Marino This file is part of the GNU MP Library.
1086d7f5d3SJohn Marino
1186d7f5d3SJohn Marino The GNU MP Library is free software; you can redistribute it and/or modify
1286d7f5d3SJohn Marino it under the terms of the GNU Lesser General Public License as published by
1386d7f5d3SJohn Marino the Free Software Foundation; either version 3 of the License, or (at your
1486d7f5d3SJohn Marino option) any later version.
1586d7f5d3SJohn Marino
1686d7f5d3SJohn Marino The GNU MP Library is distributed in the hope that it will be useful, but
1786d7f5d3SJohn Marino WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
1886d7f5d3SJohn Marino or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public
1986d7f5d3SJohn Marino License for more details.
2086d7f5d3SJohn Marino
2186d7f5d3SJohn Marino You should have received a copy of the GNU Lesser General Public License
2286d7f5d3SJohn Marino along with the GNU MP Library. If not, see http://www.gnu.org/licenses/. */
2386d7f5d3SJohn Marino
2486d7f5d3SJohn Marino
2586d7f5d3SJohn Marino #include "gmp.h"
2686d7f5d3SJohn Marino #include "gmp-impl.h"
2786d7f5d3SJohn Marino
2886d7f5d3SJohn Marino
2986d7f5d3SJohn Marino /* Calculate a remainder from {p,n} divided by 2^(GMP_NUMB_BITS*3/4)-1.
3086d7f5d3SJohn Marino The remainder is not fully reduced, it's any limb value congruent to
3186d7f5d3SJohn Marino {p,n} modulo that divisor.
3286d7f5d3SJohn Marino
3386d7f5d3SJohn Marino This implementation is only correct when GMP_NUMB_BITS is a multiple of
3486d7f5d3SJohn Marino 4.
3586d7f5d3SJohn Marino
3686d7f5d3SJohn Marino FIXME: If GMP_NAIL_BITS is some silly big value during development then
3786d7f5d3SJohn Marino it's possible the carry accumulators c0,c1,c2 could overflow.
3886d7f5d3SJohn Marino
3986d7f5d3SJohn Marino General notes:
4086d7f5d3SJohn Marino
4186d7f5d3SJohn Marino The basic idea is to use a set of N accumulators (N=3 in this case) to
4286d7f5d3SJohn Marino effectively get a remainder mod 2^(GMP_NUMB_BITS*N)-1 followed at the end
4386d7f5d3SJohn Marino by a reduction to GMP_NUMB_BITS*N/M bits (M=4 in this case) for a
4486d7f5d3SJohn Marino remainder mod 2^(GMP_NUMB_BITS*N/M)-1. N and M are chosen to give a good
4586d7f5d3SJohn Marino set of small prime factors in 2^(GMP_NUMB_BITS*N/M)-1.
4686d7f5d3SJohn Marino
4786d7f5d3SJohn Marino N=3 M=4 suits GMP_NUMB_BITS==32 and GMP_NUMB_BITS==64 quite well, giving
4886d7f5d3SJohn Marino a few more primes than a single accumulator N=1 does, and for no extra
4986d7f5d3SJohn Marino cost (assuming the processor has a decent number of registers).
5086d7f5d3SJohn Marino
5186d7f5d3SJohn Marino For strange nailified values of GMP_NUMB_BITS the idea would be to look
5286d7f5d3SJohn Marino for what N and M give good primes. With GMP_NUMB_BITS not a power of 2
5386d7f5d3SJohn Marino the choices for M may be opened up a bit. But such things are probably
5486d7f5d3SJohn Marino best done in separate code, not grafted on here. */
5586d7f5d3SJohn Marino
5686d7f5d3SJohn Marino #if GMP_NUMB_BITS % 4 == 0
5786d7f5d3SJohn Marino
5886d7f5d3SJohn Marino #define B1 (GMP_NUMB_BITS / 4)
5986d7f5d3SJohn Marino #define B2 (B1 * 2)
6086d7f5d3SJohn Marino #define B3 (B1 * 3)
6186d7f5d3SJohn Marino
6286d7f5d3SJohn Marino #define M1 ((CNST_LIMB(1) << B1) - 1)
6386d7f5d3SJohn Marino #define M2 ((CNST_LIMB(1) << B2) - 1)
6486d7f5d3SJohn Marino #define M3 ((CNST_LIMB(1) << B3) - 1)
6586d7f5d3SJohn Marino
6686d7f5d3SJohn Marino #define LOW0(n) ((n) & M3)
6786d7f5d3SJohn Marino #define HIGH0(n) ((n) >> B3)
6886d7f5d3SJohn Marino
6986d7f5d3SJohn Marino #define LOW1(n) (((n) & M2) << B1)
7086d7f5d3SJohn Marino #define HIGH1(n) ((n) >> B2)
7186d7f5d3SJohn Marino
7286d7f5d3SJohn Marino #define LOW2(n) (((n) & M1) << B2)
7386d7f5d3SJohn Marino #define HIGH2(n) ((n) >> B1)
7486d7f5d3SJohn Marino
7586d7f5d3SJohn Marino #define PARTS0(n) (LOW0(n) + HIGH0(n))
7686d7f5d3SJohn Marino #define PARTS1(n) (LOW1(n) + HIGH1(n))
7786d7f5d3SJohn Marino #define PARTS2(n) (LOW2(n) + HIGH2(n))
7886d7f5d3SJohn Marino
7986d7f5d3SJohn Marino #define ADD(c,a,val) \
8086d7f5d3SJohn Marino do { \
8186d7f5d3SJohn Marino mp_limb_t new_c; \
8286d7f5d3SJohn Marino ADDC_LIMB (new_c, a, a, val); \
8386d7f5d3SJohn Marino (c) += new_c; \
8486d7f5d3SJohn Marino } while (0)
8586d7f5d3SJohn Marino
8686d7f5d3SJohn Marino mp_limb_t
mpn_mod_34lsub1(mp_srcptr p,mp_size_t n)8786d7f5d3SJohn Marino mpn_mod_34lsub1 (mp_srcptr p, mp_size_t n)
8886d7f5d3SJohn Marino {
8986d7f5d3SJohn Marino mp_limb_t c0 = 0;
9086d7f5d3SJohn Marino mp_limb_t c1 = 0;
9186d7f5d3SJohn Marino mp_limb_t c2 = 0;
9286d7f5d3SJohn Marino mp_limb_t a0, a1, a2;
9386d7f5d3SJohn Marino
9486d7f5d3SJohn Marino ASSERT (n >= 1);
9586d7f5d3SJohn Marino ASSERT (n/3 < GMP_NUMB_MAX);
9686d7f5d3SJohn Marino
9786d7f5d3SJohn Marino a0 = a1 = a2 = 0;
9886d7f5d3SJohn Marino c0 = c1 = c2 = 0;
9986d7f5d3SJohn Marino
10086d7f5d3SJohn Marino while ((n -= 3) >= 0)
10186d7f5d3SJohn Marino {
10286d7f5d3SJohn Marino ADD (c0, a0, p[0]);
10386d7f5d3SJohn Marino ADD (c1, a1, p[1]);
10486d7f5d3SJohn Marino ADD (c2, a2, p[2]);
10586d7f5d3SJohn Marino p += 3;
10686d7f5d3SJohn Marino }
10786d7f5d3SJohn Marino
10886d7f5d3SJohn Marino if (n != -3)
10986d7f5d3SJohn Marino {
11086d7f5d3SJohn Marino ADD (c0, a0, p[0]);
11186d7f5d3SJohn Marino if (n != -2)
11286d7f5d3SJohn Marino ADD (c1, a1, p[1]);
11386d7f5d3SJohn Marino }
11486d7f5d3SJohn Marino
11586d7f5d3SJohn Marino return
11686d7f5d3SJohn Marino PARTS0 (a0) + PARTS1 (a1) + PARTS2 (a2)
11786d7f5d3SJohn Marino + PARTS1 (c0) + PARTS2 (c1) + PARTS0 (c2);
11886d7f5d3SJohn Marino }
11986d7f5d3SJohn Marino
12086d7f5d3SJohn Marino #endif
121