|
Chris@42
|
1 /*
|
|
Chris@42
|
2 * Copyright (c) 2003, 2007-14 Matteo Frigo
|
|
Chris@42
|
3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
|
|
Chris@42
|
4 *
|
|
Chris@42
|
5 * This program is free software; you can redistribute it and/or modify
|
|
Chris@42
|
6 * it under the terms of the GNU General Public License as published by
|
|
Chris@42
|
7 * the Free Software Foundation; either version 2 of the License, or
|
|
Chris@42
|
8 * (at your option) any later version.
|
|
Chris@42
|
9 *
|
|
Chris@42
|
10 * This program is distributed in the hope that it will be useful,
|
|
Chris@42
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
Chris@42
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
Chris@42
|
13 * GNU General Public License for more details.
|
|
Chris@42
|
14 *
|
|
Chris@42
|
15 * You should have received a copy of the GNU General Public License
|
|
Chris@42
|
16 * along with this program; if not, write to the Free Software
|
|
Chris@42
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
Chris@42
|
18 *
|
|
Chris@42
|
19 */
|
|
Chris@42
|
20
|
|
Chris@42
|
21
|
|
Chris@42
|
22 #include "rdft.h"
|
|
Chris@42
|
23
|
|
Chris@42
|
24 /* Check if the vecsz/sz strides are consistent with the problem
|
|
Chris@42
|
25 being in-place for vecsz.dim[vdim], or for all dimensions
|
|
Chris@42
|
26 if vdim == RNK_MINFTY. We can't just use tensor_inplace_strides
|
|
Chris@42
|
27 because rdft transforms have the unfortunate property of
|
|
Chris@42
|
28 differing input and output sizes. This routine is not
|
|
Chris@42
|
29 exhaustive; we only return 1 for the most common case. */
|
|
Chris@42
|
30 int X(rdft2_inplace_strides)(const problem_rdft2 *p, int vdim)
|
|
Chris@42
|
31 {
|
|
Chris@42
|
32 INT N, Nc;
|
|
Chris@42
|
33 INT rs, cs;
|
|
Chris@42
|
34 int i;
|
|
Chris@42
|
35
|
|
Chris@42
|
36 for (i = 0; i + 1 < p->sz->rnk; ++i)
|
|
Chris@42
|
37 if (p->sz->dims[i].is != p->sz->dims[i].os)
|
|
Chris@42
|
38 return 0;
|
|
Chris@42
|
39
|
|
Chris@42
|
40 if (!FINITE_RNK(p->vecsz->rnk) || p->vecsz->rnk == 0)
|
|
Chris@42
|
41 return 1;
|
|
Chris@42
|
42 if (!FINITE_RNK(vdim)) { /* check all vector dimensions */
|
|
Chris@42
|
43 for (vdim = 0; vdim < p->vecsz->rnk; ++vdim)
|
|
Chris@42
|
44 if (!X(rdft2_inplace_strides)(p, vdim))
|
|
Chris@42
|
45 return 0;
|
|
Chris@42
|
46 return 1;
|
|
Chris@42
|
47 }
|
|
Chris@42
|
48
|
|
Chris@42
|
49 A(vdim < p->vecsz->rnk);
|
|
Chris@42
|
50 if (p->sz->rnk == 0)
|
|
Chris@42
|
51 return(p->vecsz->dims[vdim].is == p->vecsz->dims[vdim].os);
|
|
Chris@42
|
52
|
|
Chris@42
|
53 N = X(tensor_sz)(p->sz);
|
|
Chris@42
|
54 Nc = (N / p->sz->dims[p->sz->rnk-1].n) *
|
|
Chris@42
|
55 (p->sz->dims[p->sz->rnk-1].n/2 + 1);
|
|
Chris@42
|
56 X(rdft2_strides)(p->kind, p->sz->dims + p->sz->rnk - 1, &rs, &cs);
|
|
Chris@42
|
57
|
|
Chris@42
|
58 /* the factor of 2 comes from the fact that RS is the stride
|
|
Chris@42
|
59 of p->r0 and p->r1, which is twice as large as the strides
|
|
Chris@42
|
60 in the r2r case */
|
|
Chris@42
|
61 return(p->vecsz->dims[vdim].is == p->vecsz->dims[vdim].os
|
|
Chris@42
|
62 && (X(iabs)(2 * p->vecsz->dims[vdim].os)
|
|
Chris@42
|
63 >= X(imax)(2 * Nc * X(iabs)(cs), N * X(iabs)(rs))));
|
|
Chris@42
|
64 }
|
