Chris@10
|
1 /*
|
Chris@10
|
2 * Copyright (c) 2003, 2007-11 Matteo Frigo
|
Chris@10
|
3 * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
|
Chris@10
|
4 *
|
Chris@10
|
5 * This program is free software; you can redistribute it and/or modify
|
Chris@10
|
6 * it under the terms of the GNU General Public License as published by
|
Chris@10
|
7 * the Free Software Foundation; either version 2 of the License, or
|
Chris@10
|
8 * (at your option) any later version.
|
Chris@10
|
9 *
|
Chris@10
|
10 * This program is distributed in the hope that it will be useful,
|
Chris@10
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
Chris@10
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
Chris@10
|
13 * GNU General Public License for more details.
|
Chris@10
|
14 *
|
Chris@10
|
15 * You should have received a copy of the GNU General Public License
|
Chris@10
|
16 * along with this program; if not, write to the Free Software
|
Chris@10
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
Chris@10
|
18 *
|
Chris@10
|
19 */
|
Chris@10
|
20
|
Chris@10
|
21
|
Chris@10
|
22 #include "ifftw.h"
|
Chris@10
|
23
|
Chris@10
|
24 INT X(tensor_max_index)(const tensor *sz)
|
Chris@10
|
25 {
|
Chris@10
|
26 int i;
|
Chris@10
|
27 INT ni = 0, no = 0;
|
Chris@10
|
28
|
Chris@10
|
29 A(FINITE_RNK(sz->rnk));
|
Chris@10
|
30 for (i = 0; i < sz->rnk; ++i) {
|
Chris@10
|
31 const iodim *p = sz->dims + i;
|
Chris@10
|
32 ni += (p->n - 1) * X(iabs)(p->is);
|
Chris@10
|
33 no += (p->n - 1) * X(iabs)(p->os);
|
Chris@10
|
34 }
|
Chris@10
|
35 return X(imax)(ni, no);
|
Chris@10
|
36 }
|
Chris@10
|
37
|
Chris@10
|
38 #define tensor_min_xstride(sz, xs) { \
|
Chris@10
|
39 A(FINITE_RNK(sz->rnk)); \
|
Chris@10
|
40 if (sz->rnk == 0) return 0; \
|
Chris@10
|
41 else { \
|
Chris@10
|
42 int i; \
|
Chris@10
|
43 INT s = X(iabs)(sz->dims[0].xs); \
|
Chris@10
|
44 for (i = 1; i < sz->rnk; ++i) \
|
Chris@10
|
45 s = X(imin)(s, X(iabs)(sz->dims[i].xs)); \
|
Chris@10
|
46 return s; \
|
Chris@10
|
47 } \
|
Chris@10
|
48 }
|
Chris@10
|
49
|
Chris@10
|
50 INT X(tensor_min_istride)(const tensor *sz) tensor_min_xstride(sz, is)
|
Chris@10
|
51 INT X(tensor_min_ostride)(const tensor *sz) tensor_min_xstride(sz, os)
|
Chris@10
|
52
|
Chris@10
|
53 INT X(tensor_min_stride)(const tensor *sz)
|
Chris@10
|
54 {
|
Chris@10
|
55 return X(imin)(X(tensor_min_istride)(sz), X(tensor_min_ostride)(sz));
|
Chris@10
|
56 }
|
Chris@10
|
57
|
Chris@10
|
58 int X(tensor_inplace_strides)(const tensor *sz)
|
Chris@10
|
59 {
|
Chris@10
|
60 int i;
|
Chris@10
|
61 A(FINITE_RNK(sz->rnk));
|
Chris@10
|
62 for (i = 0; i < sz->rnk; ++i) {
|
Chris@10
|
63 const iodim *p = sz->dims + i;
|
Chris@10
|
64 if (p->is != p->os)
|
Chris@10
|
65 return 0;
|
Chris@10
|
66 }
|
Chris@10
|
67 return 1;
|
Chris@10
|
68 }
|
Chris@10
|
69
|
Chris@10
|
70 int X(tensor_inplace_strides2)(const tensor *a, const tensor *b)
|
Chris@10
|
71 {
|
Chris@10
|
72 return X(tensor_inplace_strides(a)) && X(tensor_inplace_strides(b));
|
Chris@10
|
73 }
|
Chris@10
|
74
|
Chris@10
|
75 /* return true (1) iff *any* strides of sz decrease when we
|
Chris@10
|
76 tensor_inplace_copy(sz, k). */
|
Chris@10
|
77 static int tensor_strides_decrease(const tensor *sz, inplace_kind k)
|
Chris@10
|
78 {
|
Chris@10
|
79 if (FINITE_RNK(sz->rnk)) {
|
Chris@10
|
80 int i;
|
Chris@10
|
81 for (i = 0; i < sz->rnk; ++i)
|
Chris@10
|
82 if ((sz->dims[i].os - sz->dims[i].is)
|
Chris@10
|
83 * (k == INPLACE_OS ? (INT)1 : (INT)-1) < 0)
|
Chris@10
|
84 return 1;
|
Chris@10
|
85 }
|
Chris@10
|
86 return 0;
|
Chris@10
|
87 }
|
Chris@10
|
88
|
Chris@10
|
89 /* Return true (1) iff *any* strides of sz decrease when we
|
Chris@10
|
90 tensor_inplace_copy(k) *or* if *all* strides of sz are unchanged
|
Chris@10
|
91 but *any* strides of vecsz decrease. This is used in indirect.c
|
Chris@10
|
92 to determine whether to use INPLACE_IS or INPLACE_OS.
|
Chris@10
|
93
|
Chris@10
|
94 Note: X(tensor_strides_decrease)(sz, vecsz, INPLACE_IS)
|
Chris@10
|
95 || X(tensor_strides_decrease)(sz, vecsz, INPLACE_OS)
|
Chris@10
|
96 || X(tensor_inplace_strides2)(p->sz, p->vecsz)
|
Chris@10
|
97 must always be true. */
|
Chris@10
|
98 int X(tensor_strides_decrease)(const tensor *sz, const tensor *vecsz,
|
Chris@10
|
99 inplace_kind k)
|
Chris@10
|
100 {
|
Chris@10
|
101 return(tensor_strides_decrease(sz, k)
|
Chris@10
|
102 || (X(tensor_inplace_strides)(sz)
|
Chris@10
|
103 && tensor_strides_decrease(vecsz, k)));
|
Chris@10
|
104 }
|