1 /* mpn_add_err2_n -- add_n with two error terms
2 
3    Contributed by David Harvey.
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'LL CHANGE OR DISAPPEAR IN A FUTURE GNU MP RELEASE.
8 
9 Copyright 2011 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.h"
38 #include "gmp-impl.h"
39 
40 /*
41   Computes:
42 
43   (1) {rp,n} := {up,n} + {vp,n} (just like mpn_add_n) with incoming carry cy,
44   return value is carry out.
45 
46   (2) Let c[i+1] = carry from i-th limb addition (c[0] = cy).
47   Computes c[1]*yp1[n-1] + ... + c[n]*yp1[0],
48            c[1]*yp2[n-1] + ... + c[n]*yp2[0],
49   stores two-limb results at {ep,2} and {ep+2,2} respectively.
50 
51   Requires n >= 1.
52 
53   None of the outputs may overlap each other or any of the inputs, except
54   that {rp,n} may be equal to {up,n} or {vp,n}.
55 */
56 mp_limb_t
mpn_add_err2_n(mp_ptr rp,mp_srcptr up,mp_srcptr vp,mp_ptr ep,mp_srcptr yp1,mp_srcptr yp2,mp_size_t n,mp_limb_t cy)57 mpn_add_err2_n (mp_ptr rp, mp_srcptr up, mp_srcptr vp,
58                 mp_ptr ep, mp_srcptr yp1, mp_srcptr yp2,
59                 mp_size_t n, mp_limb_t cy)
60 {
61   mp_limb_t el1, eh1, el2, eh2, ul, vl, yl1, yl2, zl1, zl2, rl, sl, cy1, cy2;
62 
63   ASSERT (n >= 1);
64   ASSERT (MPN_SAME_OR_SEPARATE_P (rp, up, n));
65   ASSERT (MPN_SAME_OR_SEPARATE_P (rp, vp, n));
66   ASSERT (! MPN_OVERLAP_P (rp, n, yp1, n));
67   ASSERT (! MPN_OVERLAP_P (rp, n, yp2, n));
68   ASSERT (! MPN_OVERLAP_P (ep, 4, up, n));
69   ASSERT (! MPN_OVERLAP_P (ep, 4, vp, n));
70   ASSERT (! MPN_OVERLAP_P (ep, 4, yp1, n));
71   ASSERT (! MPN_OVERLAP_P (ep, 4, yp2, n));
72   ASSERT (! MPN_OVERLAP_P (ep, 4, rp, n));
73 
74   yp1 += n - 1;
75   yp2 += n - 1;
76   el1 = eh1 = 0;
77   el2 = eh2 = 0;
78 
79   do
80     {
81       yl1 = *yp1--;
82       yl2 = *yp2--;
83       ul = *up++;
84       vl = *vp++;
85 
86       /* ordinary add_n */
87       ADDC_LIMB (cy1, sl, ul, vl);
88       ADDC_LIMB (cy2, rl, sl, cy);
89       cy = cy1 | cy2;
90       *rp++ = rl;
91 
92       /* update (eh1:el1) */
93       zl1 = (-cy) & yl1;
94       el1 += zl1;
95       eh1 += el1 < zl1;
96 
97       /* update (eh2:el2) */
98       zl2 = (-cy) & yl2;
99       el2 += zl2;
100       eh2 += el2 < zl2;
101     }
102   while (--n);
103 
104 #if GMP_NAIL_BITS != 0
105   eh1 = (eh1 << GMP_NAIL_BITS) + (el1 >> GMP_NUMB_BITS);
106   el1 &= GMP_NUMB_MASK;
107   eh2 = (eh2 << GMP_NAIL_BITS) + (el2 >> GMP_NUMB_BITS);
108   el2 &= GMP_NUMB_MASK;
109 #endif
110 
111   ep[0] = el1;
112   ep[1] = eh1;
113   ep[2] = el2;
114   ep[3] = eh2;
115 
116   return cy;
117 }
118