1 /* mpn_toom_eval_pm2exp -- Evaluate a polynomial in +2^k and -2^k
2 
3    Contributed to the GNU project by Niels Möller
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 
38 #include "gmp.h"
39 #include "gmp-impl.h"
40 
41 /* Evaluates a polynomial of degree k > 2, in the points +2^shift and -2^shift. */
42 int
mpn_toom_eval_pm2exp(mp_ptr xp2,mp_ptr xm2,unsigned k,mp_srcptr xp,mp_size_t n,mp_size_t hn,unsigned shift,mp_ptr tp)43 mpn_toom_eval_pm2exp (mp_ptr xp2, mp_ptr xm2, unsigned k,
44 		      mp_srcptr xp, mp_size_t n, mp_size_t hn, unsigned shift,
45 		      mp_ptr tp)
46 {
47   unsigned i;
48   int neg;
49 #if HAVE_NATIVE_mpn_addlsh_n
50   mp_limb_t cy;
51 #endif
52 
53   ASSERT (k >= 3);
54   ASSERT (shift*k < GMP_NUMB_BITS);
55 
56   ASSERT (hn > 0);
57   ASSERT (hn <= n);
58 
59   /* The degree k is also the number of full-size coefficients, so
60    * that last coefficient, of size hn, starts at xp + k*n. */
61 
62 #if HAVE_NATIVE_mpn_addlsh_n
63   xp2[n] = mpn_addlsh_n (xp2, xp, xp + 2*n, n, 2*shift);
64   for (i = 4; i < k; i += 2)
65     xp2[n] += mpn_addlsh_n (xp2, xp2, xp + i*n, n, i*shift);
66 
67   tp[n] = mpn_lshift (tp, xp+n, n, shift);
68   for (i = 3; i < k; i+= 2)
69     tp[n] += mpn_addlsh_n (tp, tp, xp+i*n, n, i*shift);
70 
71   if (k & 1)
72     {
73       cy = mpn_addlsh_n (tp, tp, xp+k*n, hn, k*shift);
74       MPN_INCR_U (tp + hn, n+1 - hn, cy);
75     }
76   else
77     {
78       cy = mpn_addlsh_n (xp2, xp2, xp+k*n, hn, k*shift);
79       MPN_INCR_U (xp2 + hn, n+1 - hn, cy);
80     }
81 
82 #else /* !HAVE_NATIVE_mpn_addlsh_n */
83   xp2[n] = mpn_lshift (tp, xp+2*n, n, 2*shift);
84   xp2[n] += mpn_add_n (xp2, xp, tp, n);
85   for (i = 4; i < k; i += 2)
86     {
87       xp2[n] += mpn_lshift (tp, xp + i*n, n, i*shift);
88       xp2[n] += mpn_add_n (xp2, xp2, tp, n);
89     }
90 
91   tp[n] = mpn_lshift (tp, xp+n, n, shift);
92   for (i = 3; i < k; i+= 2)
93     {
94       tp[n] += mpn_lshift (xm2, xp + i*n, n, i*shift);
95       tp[n] += mpn_add_n (tp, tp, xm2, n);
96     }
97 
98   xm2[hn] = mpn_lshift (xm2, xp + k*n, hn, k*shift);
99   if (k & 1)
100     mpn_add (tp, tp, n+1, xm2, hn+1);
101   else
102     mpn_add (xp2, xp2, n+1, xm2, hn+1);
103 #endif /* !HAVE_NATIVE_mpn_addlsh_n */
104 
105   neg = (mpn_cmp (xp2, tp, n + 1) < 0) ? ~0 : 0;
106 
107 #if HAVE_NATIVE_mpn_add_n_sub_n
108   if (neg)
109     mpn_add_n_sub_n (xp2, xm2, tp, xp2, n + 1);
110   else
111     mpn_add_n_sub_n (xp2, xm2, xp2, tp, n + 1);
112 #else /* !HAVE_NATIVE_mpn_add_n_sub_n */
113   if (neg)
114     mpn_sub_n (xm2, tp, xp2, n + 1);
115   else
116     mpn_sub_n (xm2, xp2, tp, n + 1);
117 
118   mpn_add_n (xp2, xp2, tp, n + 1);
119 #endif /* !HAVE_NATIVE_mpn_add_n_sub_n */
120 
121   /* FIXME: the following asserts are useless if (k+1)*shift >= GMP_LIMB_BITS */
122   ASSERT ((k+1)*shift >= GMP_LIMB_BITS ||
123 	  xp2[n] < ((CNST_LIMB(1)<<((k+1)*shift))-1)/((CNST_LIMB(1)<<shift)-1));
124   ASSERT ((k+2)*shift >= GMP_LIMB_BITS ||
125 	  xm2[n] < ((CNST_LIMB(1)<<((k+2)*shift))-((k&1)?(CNST_LIMB(1)<<shift):1))/((CNST_LIMB(1)<<(2*shift))-1));
126 
127   return neg;
128 }
129