xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/generic/toom_eval_pm2.c (revision 72c7faa4dbb41dbb0238d6b4a109da0d4b236dd4)
1 /* mpn_toom_eval_pm2 -- Evaluate a polynomial in +2 and -2
2 
3    Contributed to the GNU project by Niels Möller and Marco Bodrato
4 
5    THE FUNCTION IN THIS FILE IS INTERNAL WITH A MUTABLE INTERFACE.  IT IS ONLY
6    SAFE TO REACH IT THROUGH DOCUMENTED INTERFACES.  IN FACT, IT IS ALMOST
7    GUARANTEED THAT IT WILL CHANGE OR DISAPPEAR IN A FUTURE GNU MP RELEASE.
8 
9 Copyright 2009 Free Software Foundation, Inc.
10 
11 This file is part of the GNU MP Library.
12 
13 The GNU MP Library is free software; you can redistribute it and/or modify
14 it under the terms of either:
15 
16   * the GNU Lesser General Public License as published by the Free
17     Software Foundation; either version 3 of the License, or (at your
18     option) any later version.
19 
20 or
21 
22   * the GNU General Public License as published by the Free Software
23     Foundation; either version 2 of the License, or (at your option) any
24     later version.
25 
26 or both in parallel, as here.
27 
28 The GNU MP Library is distributed in the hope that it will be useful, but
29 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
30 or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
31 for more details.
32 
33 You should have received copies of the GNU General Public License and the
34 GNU Lesser General Public License along with the GNU MP Library.  If not,
35 see https://www.gnu.org/licenses/.  */
36 
37 #include "gmp-impl.h"
38 
39 /* DO_addlsh2(d,a,b,n,cy) computes cy,{d,n} <- {a,n} + 4*(cy,{b,n}), it
40    can be used as DO_addlsh2(d,a,d,n,d[n]), for accumulation on {d,n+1}. */
41 #if HAVE_NATIVE_mpn_addlsh2_n
42 #define DO_addlsh2(d, a, b, n, cy)	\
43 do {					\
44   (cy) <<= 2;				\
45   (cy) += mpn_addlsh2_n(d, a, b, n);	\
46 } while (0)
47 #else
48 #if HAVE_NATIVE_mpn_addlsh_n
49 #define DO_addlsh2(d, a, b, n, cy)	\
50 do {					\
51   (cy) <<= 2;				\
52   (cy) += mpn_addlsh_n(d, a, b, n, 2);	\
53 } while (0)
54 #else
55 /* The following is not a general substitute for addlsh2.
56    It is correct if d == b, but it is not if d == a.  */
57 #define DO_addlsh2(d, a, b, n, cy)	\
58 do {					\
59   (cy) <<= 2;				\
60   (cy) += mpn_lshift(d, b, n, 2);	\
61   (cy) += mpn_add_n(d, d, a, n);	\
62 } while (0)
63 #endif
64 #endif
65 
66 /* Evaluates a polynomial of degree 2 < k < GMP_NUMB_BITS, in the
67    points +2 and -2. */
68 int
mpn_toom_eval_pm2(mp_ptr xp2,mp_ptr xm2,unsigned k,mp_srcptr xp,mp_size_t n,mp_size_t hn,mp_ptr tp)69 mpn_toom_eval_pm2 (mp_ptr xp2, mp_ptr xm2, unsigned k,
70 		   mp_srcptr xp, mp_size_t n, mp_size_t hn, mp_ptr tp)
71 {
72   int i;
73   int neg;
74   mp_limb_t cy;
75 
76   ASSERT (k >= 3);
77   ASSERT (k < GMP_NUMB_BITS);
78 
79   ASSERT (hn > 0);
80   ASSERT (hn <= n);
81 
82   /* The degree k is also the number of full-size coefficients, so
83    * that last coefficient, of size hn, starts at xp + k*n. */
84 
85   cy = 0;
86   DO_addlsh2 (xp2, xp + (k-2) * n, xp + k * n, hn, cy);
87   if (hn != n)
88     cy = mpn_add_1 (xp2 + hn, xp + (k-2) * n + hn, n - hn, cy);
89   for (i = k - 4; i >= 0; i -= 2)
90     DO_addlsh2 (xp2, xp + i * n, xp2, n, cy);
91   xp2[n] = cy;
92 
93   k--;
94 
95   cy = 0;
96   DO_addlsh2 (tp, xp + (k-2) * n, xp + k * n, n, cy);
97   for (i = k - 4; i >= 0; i -= 2)
98     DO_addlsh2 (tp, xp + i * n, tp, n, cy);
99   tp[n] = cy;
100 
101   if (k & 1)
102     ASSERT_NOCARRY(mpn_lshift (tp , tp , n + 1, 1));
103   else
104     ASSERT_NOCARRY(mpn_lshift (xp2, xp2, n + 1, 1));
105 
106   neg = (mpn_cmp (xp2, tp, n + 1) < 0) ? ~0 : 0;
107 
108 #if HAVE_NATIVE_mpn_add_n_sub_n
109   if (neg)
110     mpn_add_n_sub_n (xp2, xm2, tp, xp2, n + 1);
111   else
112     mpn_add_n_sub_n (xp2, xm2, xp2, tp, n + 1);
113 #else /* !HAVE_NATIVE_mpn_add_n_sub_n */
114   if (neg)
115     mpn_sub_n (xm2, tp, xp2, n + 1);
116   else
117     mpn_sub_n (xm2, xp2, tp, n + 1);
118 
119   mpn_add_n (xp2, xp2, tp, n + 1);
120 #endif /* !HAVE_NATIVE_mpn_add_n_sub_n */
121 
122   ASSERT (xp2[n] < (1<<(k+2))-1);
123   ASSERT (xm2[n] < ((1<<(k+3))-1 - (1^k&1))/3);
124 
125   neg ^= ((k & 1) - 1);
126 
127   return neg;
128 }
129 
130 #undef DO_addlsh2
131