xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/generic/brootinv.c (revision 72c7faa4dbb41dbb0238d6b4a109da0d4b236dd4)
1 /* mpn_brootinv, compute r such that r^k * y = 1 (mod 2^b).
2 
3    Contributed to the GNU project by Martin Boij (as part of perfpow.c).
4 
5 Copyright 2009, 2010, 2012, 2013, 2018 Free Software Foundation, Inc.
6 
7 This file is part of the GNU MP Library.
8 
9 The GNU MP Library is free software; you can redistribute it and/or modify
10 it under the terms of either:
11 
12   * the GNU Lesser General Public License as published by the Free
13     Software Foundation; either version 3 of the License, or (at your
14     option) any later version.
15 
16 or
17 
18   * the GNU General Public License as published by the Free Software
19     Foundation; either version 2 of the License, or (at your option) any
20     later version.
21 
22 or both in parallel, as here.
23 
24 The GNU MP Library is distributed in the hope that it will be useful, but
25 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
26 or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
27 for more details.
28 
29 You should have received copies of the GNU General Public License and the
30 GNU Lesser General Public License along with the GNU MP Library.  If not,
31 see https://www.gnu.org/licenses/.  */
32 
33 #include "gmp-impl.h"
34 
35 /* Computes a^2e (mod B). Uses right-to-left binary algorithm, since
36    typical use will have e small. */
37 static mp_limb_t
powsquaredlimb(mp_limb_t a,mp_limb_t e)38 powsquaredlimb (mp_limb_t a, mp_limb_t e)
39 {
40   mp_limb_t r;
41 
42   r = 1;
43   /* if (LIKELY (e != 0)) */
44   do {
45     a *= a;
46     if (e & 1)
47       r *= a;
48     e >>= 1;
49   } while (e != 0);
50 
51   return r;
52 }
53 
54 /* Compute r such that r^k * y = 1 (mod B^n).
55 
56    Iterates
57      r' <-- k^{-1} ((k+1) r - r^{k+1} y) (mod 2^b)
58    using Hensel lifting, each time doubling the number of known bits in r.
59 
60    Works just for odd k.  Else the Hensel lifting degenerates.
61 
62    FIXME:
63 
64      (1) Make it work for k == GMP_LIMB_MAX (k+1 below overflows).
65 
66      (2) Rewrite iteration as
67 	   r' <-- r - k^{-1} r (r^k y - 1)
68 	 and take advantage of the zero low part of r^k y - 1.
69 
70      (3) Use wrap-around trick.
71 
72      (4) Use a small table to get starting value.
73 
74    Scratch need: bn + (((bn + 1) >> 1) + 1) + scratch for mpn_powlo
75    Currently mpn_powlo requires 3*bn
76    so that 5*bn is surely enough, where bn = ceil (bnb / GMP_NUMB_BITS).
77 */
78 
79 void
mpn_brootinv(mp_ptr rp,mp_srcptr yp,mp_size_t bn,mp_limb_t k,mp_ptr tp)80 mpn_brootinv (mp_ptr rp, mp_srcptr yp, mp_size_t bn, mp_limb_t k, mp_ptr tp)
81 {
82   mp_ptr tp2, tp3;
83   mp_limb_t kinv, k2, r0, y0;
84   mp_size_t order[GMP_LIMB_BITS + 1];
85   int d;
86 
87   ASSERT (bn > 0);
88   ASSERT ((k & 1) != 0);
89 
90   tp2 = tp + bn;
91   tp3 = tp + bn + ((bn + 3) >> 1);
92   k2 = (k >> 1) + 1; /* (k + 1) / 2 , but avoid k+1 overflow */
93 
94   binvert_limb (kinv, k);
95 
96   /* 4-bit initial approximation:
97 
98    y%16 | 1  3  5  7  9 11 13 15,
99     k%4 +-------------------------+k2%2
100      1  | 1 11 13  7  9  3  5 15  |  1
101      3  | 1  3  5  7  9 11 13 15  |  0
102 
103   */
104   y0 = yp[0];
105 
106   r0 = y0 ^ (((y0 << 1) ^ (y0 << 2)) & (k2 << 3) & 8);			/* 4 bits */
107   r0 = kinv * (k2 * r0 * 2 - y0 * powsquaredlimb(r0, k2 & 0x3f));	/* 8 bits */
108   r0 = kinv * (k2 * r0 * 2 - y0 * powsquaredlimb(r0, k2 & 0x3fff));	/* 16 bits */
109 #if GMP_NUMB_BITS > 16
110   {
111     unsigned prec = 16;
112     do
113       {
114 	r0 = kinv * (k2 * r0 * 2 - y0 * powsquaredlimb(r0, k2));
115 	prec *= 2;
116       }
117     while (prec < GMP_NUMB_BITS);
118   }
119 #endif
120 
121   rp[0] = r0;
122   if (bn == 1)
123     return;
124 
125   d = 0;
126   for (; bn != 2; bn = (bn + 1) >> 1)
127     order[d++] = bn;
128 
129   order[d] = 2;
130   bn = 1;
131 
132   do
133     {
134       mpn_sqr (tp, rp, bn); /* Result may overlap tp2 */
135       tp2[bn] = mpn_mul_1 (tp2, rp, bn, k2 << 1);
136 
137       bn = order[d];
138 
139       mpn_powlo (rp, tp, &k2, 1, bn, tp3);
140       mpn_mullo_n (tp, yp, rp, bn);
141 
142       /* mpn_sub (tp, tp2, ((bn + 1) >> 1) + 1, tp, bn); */
143       /* The function above is not handled, ((bn + 1) >> 1) + 1 <= bn*/
144       {
145 	mp_size_t pbn = (bn + 3) >> 1; /* Size of tp2 */
146 	int borrow;
147 	borrow = mpn_sub_n (tp, tp2, tp, pbn) != 0;
148 	if (bn > pbn) /* 3 < bn */
149 	  {
150 	    if (borrow)
151 	      mpn_com (tp + pbn, tp + pbn, bn - pbn);
152 	    else
153 	      mpn_neg (tp + pbn, tp + pbn, bn - pbn);
154 	  }
155       }
156       mpn_pi1_bdiv_q_1 (rp, tp, bn, k, kinv, 0);
157     }
158   while (--d >= 0);
159 }
160