xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/generic/mul_basecase.c (revision f89f6560d453f5e37386cc7938c072d2f528b9fa)
1 /* mpn_mul_basecase -- Internal routine to multiply two natural numbers
2    of length m and n.
3 
4    THIS IS AN INTERNAL FUNCTION WITH A MUTABLE INTERFACE.  IT IS ONLY
5    SAFE TO REACH THIS FUNCTION THROUGH DOCUMENTED INTERFACES.
6 
7 Copyright 1991, 1992, 1993, 1994, 1996, 1997, 2000, 2001, 2002 Free Software
8 Foundation, Inc.
9 
10 This file is part of the GNU MP Library.
11 
12 The GNU MP Library is free software; you can redistribute it and/or modify
13 it under the terms of the GNU Lesser General Public License as published by
14 the Free Software Foundation; either version 3 of the License, or (at your
15 option) any later version.
16 
17 The GNU MP Library is distributed in the hope that it will be useful, but
18 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
19 or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
20 License for more details.
21 
22 You should have received a copy of the GNU Lesser General Public License
23 along with the GNU MP Library.  If not, see http://www.gnu.org/licenses/.  */
24 
25 #include "gmp.h"
26 #include "gmp-impl.h"
27 
28 
29 /* Multiply {up,usize} by {vp,vsize} and write the result to
30    {prodp,usize+vsize}.  Must have usize>=vsize.
31 
32    Note that prodp gets usize+vsize limbs stored, even if the actual result
33    only needs usize+vsize-1.
34 
35    There's no good reason to call here with vsize>=MUL_TOOM22_THRESHOLD.
36    Currently this is allowed, but it might not be in the future.
37 
38    This is the most critical code for multiplication.  All multiplies rely
39    on this, both small and huge.  Small ones arrive here immediately, huge
40    ones arrive here as this is the base case for Karatsuba's recursive
41    algorithm.  */
42 
43 void
44 mpn_mul_basecase (mp_ptr rp,
45 		  mp_srcptr up, mp_size_t un,
46 		  mp_srcptr vp, mp_size_t vn)
47 {
48   ASSERT (un >= vn);
49   ASSERT (vn >= 1);
50   ASSERT (! MPN_OVERLAP_P (rp, un+vn, up, un));
51   ASSERT (! MPN_OVERLAP_P (rp, un+vn, vp, vn));
52 
53   /* We first multiply by the low order limb (or depending on optional function
54      availability, limbs).  This result can be stored, not added, to rp.  We
55      also avoid a loop for zeroing this way.  */
56 
57 #if HAVE_NATIVE_mpn_mul_2
58   if (vn >= 2)
59     {
60       rp[un + 1] = mpn_mul_2 (rp, up, un, vp);
61       rp += 2, vp += 2, vn -= 2;
62     }
63   else
64     {
65       rp[un] = mpn_mul_1 (rp, up, un, vp[0]);
66       return;
67     }
68 #else
69   rp[un] = mpn_mul_1 (rp, up, un, vp[0]);
70   rp += 1, vp += 1, vn -= 1;
71 #endif
72 
73   /* Now accumulate the product of up[] and the next higher limb (or depending
74      on optional function availability, limbs) from vp[].  */
75 
76 #define MAX_LEFT MP_SIZE_T_MAX	/* Used to simplify loops into if statements */
77 
78 
79 #if HAVE_NATIVE_mpn_addmul_6
80   while (vn >= 6)
81     {
82       rp[un + 6 - 1] = mpn_addmul_6 (rp, up, un, vp);
83       if (MAX_LEFT == 6)
84 	return;
85       rp += 6, vp += 6, vn -= 6;
86       if (MAX_LEFT < 2 * 6)
87 	break;
88     }
89 #undef MAX_LEFT
90 #define MAX_LEFT (6 - 1)
91 #endif
92 
93 #if HAVE_NATIVE_mpn_addmul_5
94   while (vn >= 5)
95     {
96       rp[un + 5 - 1] = mpn_addmul_5 (rp, up, un, vp);
97       if (MAX_LEFT == 5)
98 	return;
99       rp += 5, vp += 5, vn -= 5;
100       if (MAX_LEFT < 2 * 5)
101 	break;
102     }
103 #undef MAX_LEFT
104 #define MAX_LEFT (5 - 1)
105 #endif
106 
107 #if HAVE_NATIVE_mpn_addmul_4
108   while (vn >= 4)
109     {
110       rp[un + 4 - 1] = mpn_addmul_4 (rp, up, un, vp);
111       if (MAX_LEFT == 4)
112 	return;
113       rp += 4, vp += 4, vn -= 4;
114       if (MAX_LEFT < 2 * 4)
115 	break;
116     }
117 #undef MAX_LEFT
118 #define MAX_LEFT (4 - 1)
119 #endif
120 
121 #if HAVE_NATIVE_mpn_addmul_3
122   while (vn >= 3)
123     {
124       rp[un + 3 - 1] = mpn_addmul_3 (rp, up, un, vp);
125       if (MAX_LEFT == 3)
126 	return;
127       rp += 3, vp += 3, vn -= 3;
128       if (MAX_LEFT < 2 * 3)
129 	break;
130     }
131 #undef MAX_LEFT
132 #define MAX_LEFT (3 - 1)
133 #endif
134 
135 #if HAVE_NATIVE_mpn_addmul_2
136   while (vn >= 2)
137     {
138       rp[un + 2 - 1] = mpn_addmul_2 (rp, up, un, vp);
139       if (MAX_LEFT == 2)
140 	return;
141       rp += 2, vp += 2, vn -= 2;
142       if (MAX_LEFT < 2 * 2)
143 	break;
144     }
145 #undef MAX_LEFT
146 #define MAX_LEFT (2 - 1)
147 #endif
148 
149   while (vn >= 1)
150     {
151       rp[un] = mpn_addmul_1 (rp, up, un, vp[0]);
152       if (MAX_LEFT == 1)
153 	return;
154       rp += 1, vp += 1, vn -= 1;
155     }
156 }
157