|
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 /* out of place 2D copy routines */
|
|
Chris@82
|
22 #include "kernel/ifftw.h"
|
|
Chris@82
|
23
|
|
Chris@82
|
24 void X(tile2d)(INT n0l, INT n0u, INT n1l, INT n1u, INT tilesz,
|
|
Chris@82
|
25 void (*f)(INT n0l, INT n0u, INT n1l, INT n1u, void *args),
|
|
Chris@82
|
26 void *args)
|
|
Chris@82
|
27 {
|
|
Chris@82
|
28 INT d0, d1;
|
|
Chris@82
|
29
|
|
Chris@82
|
30 A(tilesz > 0); /* infinite loops otherwise */
|
|
Chris@82
|
31
|
|
Chris@82
|
32 tail:
|
|
Chris@82
|
33 d0 = n0u - n0l;
|
|
Chris@82
|
34 d1 = n1u - n1l;
|
|
Chris@82
|
35
|
|
Chris@82
|
36 if (d0 >= d1 && d0 > tilesz) {
|
|
Chris@82
|
37 INT n0m = (n0u + n0l) / 2;
|
|
Chris@82
|
38 X(tile2d)(n0l, n0m, n1l, n1u, tilesz, f, args);
|
|
Chris@82
|
39 n0l = n0m; goto tail;
|
|
Chris@82
|
40 } else if (/* d1 >= d0 && */ d1 > tilesz) {
|
|
Chris@82
|
41 INT n1m = (n1u + n1l) / 2;
|
|
Chris@82
|
42 X(tile2d)(n0l, n0u, n1l, n1m, tilesz, f, args);
|
|
Chris@82
|
43 n1l = n1m; goto tail;
|
|
Chris@82
|
44 } else {
|
|
Chris@82
|
45 f(n0l, n0u, n1l, n1u, args);
|
|
Chris@82
|
46 }
|
|
Chris@82
|
47 }
|
|
Chris@82
|
48
|
|
Chris@82
|
49 INT X(compute_tilesz)(INT vl, int how_many_tiles_in_cache)
|
|
Chris@82
|
50 {
|
|
Chris@82
|
51 return X(isqrt)(CACHESIZE /
|
|
Chris@82
|
52 (((INT)sizeof(R)) * vl * (INT)how_many_tiles_in_cache));
|
|
Chris@82
|
53 }
|
