1 /* mpn_toom_eval_pm2exp -- Evaluate a polynomial in +2^k and -2^k
2 
3    Contributed to the GNU project by Niels M�ller
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 the GNU Lesser General Public License as published by
15 the Free Software Foundation; either version 3 of the License, or (at your
16 option) any later version.
17 
18 The GNU MP Library is distributed in the hope that it will be useful, but
19 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
20 or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
21 License for more details.
22 
23 You should have received a copy of the GNU Lesser General Public License
24 along with the GNU MP Library.  If not, see http://www.gnu.org/licenses/.  */
25 
26 
27 #include "gmp.h"
28 #include "gmp-impl.h"
29 
30 /* Evaluates a polynomial of degree k > 2, in the points +2^shift and -2^shift. */
31 int
32 mpn_toom_eval_pm2exp (mp_ptr xp2, mp_ptr xm2, unsigned k,
33 		      mp_srcptr xp, mp_size_t n, mp_size_t hn, unsigned shift,
34 		      mp_ptr tp)
35 {
36   unsigned i;
37   int neg;
38 #if HAVE_NATIVE_mpn_addlsh_n
39   mp_limb_t cy;
40 #endif
41 
42   ASSERT (k >= 3);
43   ASSERT (shift*k < GMP_NUMB_BITS);
44 
45   ASSERT (hn > 0);
46   ASSERT (hn <= n);
47 
48   /* The degree k is also the number of full-size coefficients, so
49    * that last coefficient, of size hn, starts at xp + k*n. */
50 
51 #if HAVE_NATIVE_mpn_addlsh_n
52   xp2[n] = mpn_addlsh_n (xp2, xp, xp + 2*n, n, 2*shift);
53   for (i = 4; i < k; i += 2)
54     xp2[n] += mpn_addlsh_n (xp2, xp2, xp + i*n, n, i*shift);
55 
56   tp[n] = mpn_lshift (tp, xp+n, n, shift);
57   for (i = 3; i < k; i+= 2)
58     tp[n] += mpn_addlsh_n (tp, tp, xp+i*n, n, i*shift);
59 
60   if (k & 1)
61     {
62       cy = mpn_addlsh_n (tp, tp, xp+k*n, hn, k*shift);
63       MPN_INCR_U (tp + hn, n+1 - hn, cy);
64     }
65   else
66     {
67       cy = mpn_addlsh_n (xp2, xp2, xp+k*n, hn, k*shift);
68       MPN_INCR_U (xp2 + hn, n+1 - hn, cy);
69     }
70 
71 #else /* !HAVE_NATIVE_mpn_addlsh_n */
72   xp2[n] = mpn_lshift (tp, xp+2*n, n, 2*shift);
73   xp2[n] += mpn_add_n (xp2, xp, tp, n);
74   for (i = 4; i < k; i += 2)
75     {
76       xp2[n] += mpn_lshift (tp, xp + i*n, n, i*shift);
77       xp2[n] += mpn_add_n (xp2, xp2, tp, n);
78     }
79 
80   tp[n] = mpn_lshift (tp, xp+n, n, shift);
81   for (i = 3; i < k; i+= 2)
82     {
83       tp[n] += mpn_lshift (xm2, xp + i*n, n, i*shift);
84       tp[n] += mpn_add_n (tp, tp, xm2, n);
85     }
86 
87   xm2[hn] = mpn_lshift (xm2, xp + k*n, hn, k*shift);
88   if (k & 1)
89     mpn_add (tp, tp, n+1, xm2, hn+1);
90   else
91     mpn_add (xp2, xp2, n+1, xm2, hn+1);
92 #endif /* !HAVE_NATIVE_mpn_addlsh_n */
93 
94   neg = (mpn_cmp (xp2, tp, n + 1) < 0) ? ~0 : 0;
95 
96 #if HAVE_NATIVE_mpn_add_n_sub_n
97   if (neg)
98     mpn_add_n_sub_n (xp2, xm2, tp, xp2, n + 1);
99   else
100     mpn_add_n_sub_n (xp2, xm2, xp2, tp, n + 1);
101 #else /* !HAVE_NATIVE_mpn_add_n_sub_n */
102   if (neg)
103     mpn_sub_n (xm2, tp, xp2, n + 1);
104   else
105     mpn_sub_n (xm2, xp2, tp, n + 1);
106 
107   mpn_add_n (xp2, xp2, tp, n + 1);
108 #endif /* !HAVE_NATIVE_mpn_add_n_sub_n */
109 
110   /* FIXME: the following asserts are useless if (k+1)*shift >= GMP_LIMB_BITS */
111   ASSERT ((k+1)*shift >= GMP_LIMB_BITS ||
112 	  xp2[n] < ((CNST_LIMB(1)<<((k+1)*shift))-1)/((CNST_LIMB(1)<<shift)-1));
113   ASSERT ((k+2)*shift >= GMP_LIMB_BITS ||
114 	  xm2[n] < ((CNST_LIMB(1)<<((k+2)*shift))-((k&1)?(CNST_LIMB(1)<<shift):1))/((CNST_LIMB(1)<<(2*shift))-1));
115 
116   return neg;
117 }
118