|
Chris@82
|
1 /*
|
|
Chris@82
|
2 * Copyright (c) 2003, 2007-14 Matteo Frigo
|
|
Chris@82
|
3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
|
|
Chris@82
|
4 *
|
|
Chris@82
|
5 * This program is free software; you can redistribute it and/or modify
|
|
Chris@82
|
6 * it under the terms of the GNU General Public License as published by
|
|
Chris@82
|
7 * the Free Software Foundation; either version 2 of the License, or
|
|
Chris@82
|
8 * (at your option) any later version.
|
|
Chris@82
|
9 *
|
|
Chris@82
|
10 * This program is distributed in the hope that it will be useful,
|
|
Chris@82
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
Chris@82
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
Chris@82
|
13 * GNU General Public License for more details.
|
|
Chris@82
|
14 *
|
|
Chris@82
|
15 * You should have received a copy of the GNU General Public License
|
|
Chris@82
|
16 * along with this program; if not, write to the Free Software
|
|
Chris@82
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
Chris@82
|
18 *
|
|
Chris@82
|
19 */
|
|
Chris@82
|
20
|
|
Chris@82
|
21 /* routines shared by the various buffered solvers */
|
|
Chris@82
|
22
|
|
Chris@82
|
23 #include "kernel/ifftw.h"
|
|
Chris@82
|
24
|
|
Chris@82
|
25 #define DEFAULT_MAXNBUF ((INT)256)
|
|
Chris@82
|
26
|
|
Chris@82
|
27 /* approx. 512KB of buffers for complex data */
|
|
Chris@82
|
28 #define MAXBUFSZ (256 * 1024 / (INT)(sizeof(R)))
|
|
Chris@82
|
29
|
|
Chris@82
|
30 INT X(nbuf)(INT n, INT vl, INT maxnbuf)
|
|
Chris@82
|
31 {
|
|
Chris@82
|
32 INT i, nbuf, lb;
|
|
Chris@82
|
33
|
|
Chris@82
|
34 if (!maxnbuf)
|
|
Chris@82
|
35 maxnbuf = DEFAULT_MAXNBUF;
|
|
Chris@82
|
36
|
|
Chris@82
|
37 nbuf = X(imin)(maxnbuf,
|
|
Chris@82
|
38 X(imin)(vl, X(imax)((INT)1, MAXBUFSZ / n)));
|
|
Chris@82
|
39
|
|
Chris@82
|
40 /*
|
|
Chris@82
|
41 * Look for a buffer number (not too small) that divides the
|
|
Chris@82
|
42 * vector length, in order that we only need one child plan:
|
|
Chris@82
|
43 */
|
|
Chris@82
|
44 lb = X(imax)(1, nbuf / 4);
|
|
Chris@82
|
45 for (i = nbuf; i >= lb; --i)
|
|
Chris@82
|
46 if (vl % i == 0)
|
|
Chris@82
|
47 return i;
|
|
Chris@82
|
48
|
|
Chris@82
|
49 /* whatever... */
|
|
Chris@82
|
50 return nbuf;
|
|
Chris@82
|
51 }
|
|
Chris@82
|
52
|
|
Chris@82
|
53 #define SKEW 6 /* need to be even for SIMD */
|
|
Chris@82
|
54 #define SKEWMOD 8
|
|
Chris@82
|
55
|
|
Chris@82
|
56 INT X(bufdist)(INT n, INT vl)
|
|
Chris@82
|
57 {
|
|
Chris@82
|
58 if (vl == 1)
|
|
Chris@82
|
59 return n;
|
|
Chris@82
|
60 else
|
|
Chris@82
|
61 /* return smallest X such that X >= N and X == SKEW (mod SKEWMOD) */
|
|
Chris@82
|
62 return n + X(modulo)(SKEW - n, SKEWMOD);
|
|
Chris@82
|
63 }
|
|
Chris@82
|
64
|
|
Chris@82
|
65 int X(toobig)(INT n)
|
|
Chris@82
|
66 {
|
|
Chris@82
|
67 return n > MAXBUFSZ;
|
|
Chris@82
|
68 }
|
|
Chris@82
|
69
|
|
Chris@82
|
70 /* TRUE if there exists i < which such that maxnbuf[i] and
|
|
Chris@82
|
71 maxnbuf[which] yield the same value, in which case we canonicalize
|
|
Chris@82
|
72 on the minimum value */
|
|
Chris@82
|
73 int X(nbuf_redundant)(INT n, INT vl, size_t which,
|
|
Chris@82
|
74 const INT *maxnbuf, size_t nmaxnbuf)
|
|
Chris@82
|
75 {
|
|
Chris@82
|
76 size_t i;
|
|
Chris@82
|
77 (void)nmaxnbuf; /* UNUSED */
|
|
Chris@82
|
78 for (i = 0; i < which; ++i)
|
|
Chris@82
|
79 if (X(nbuf)(n, vl, maxnbuf[i]) == X(nbuf)(n, vl, maxnbuf[which]))
|
|
Chris@82
|
80 return 1;
|
|
Chris@82
|
81 return 0;
|
|
Chris@82
|
82 }
|
