1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
|
/*
* Copyright (c) 2003, 2006 Matteo Frigo
* Copyright (c) 2003, 2006 Massachusetts Institute of Technology
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
*/
/* $Id: tensor7.c,v 1.10 2006-01-05 03:04:27 stevenj Exp $ */
#include "ifftw.h"
static int signof(INT x)
{
if (x < 0) return -1;
if (x == 0) return 0;
/* if (x > 0) */ return 1;
}
/* total order among iodim's */
int X(dimcmp)(const iodim *a, const iodim *b)
{
INT sai = X(iabs)(a->is), sbi = X(iabs)(b->is);
INT sao = X(iabs)(a->os), sbo = X(iabs)(b->os);
INT sam = X(imin)(sai, sao), sbm = X(imin)(sbi, sbo);
/* in descending order of min{istride, ostride} */
if (sam != sbm)
return signof(sbm - sam);
/* in case of a tie, in descending order of istride */
if (sbi != sai)
return signof(sbi - sai);
/* in case of a tie, in descending order of ostride */
if (sbo != sao)
return signof(sbo - sao);
/* in case of a tie, in ascending order of n */
return signof(a->n - b->n);
}
/* Like tensor_copy, but eliminate n == 1 dimensions, which
never affect any transform or transform vector.
Also, we sort the tensor into a canonical order of decreasing
is. In general, processing a loop/array in order of
decreasing stride will improve locality; sorting also makes the
analysis in fftw_tensor_contiguous (below) easier. The choice
of is over os is mostly arbitrary, and hopefully
shouldn't affect things much. Normally, either the os will be
in the same order as is (for e.g. multi-dimensional
transforms) or will be in opposite order (e.g. for Cooley-Tukey
recursion). (Both forward and backwards traversal of the tensor
are considered e.g. by vrank-geq1, so sorting in increasing
vs. decreasing order is not really important.) */
tensor *X(tensor_compress)(const tensor *sz)
{
int i, rnk;
tensor *x;
A(FINITE_RNK(sz->rnk));
for (i = rnk = 0; i < sz->rnk; ++i) {
A(sz->dims[i].n > 0);
if (sz->dims[i].n != 1)
++rnk;
}
x = X(mktensor)(rnk);
for (i = rnk = 0; i < sz->rnk; ++i) {
if (sz->dims[i].n != 1)
x->dims[rnk++] = sz->dims[i];
}
if (rnk) {
/* God knows how qsort() behaves if n==0 */
qsort(x->dims, (size_t)x->rnk, sizeof(iodim),
(int (*)(const void *, const void *))X(dimcmp));
}
return x;
}
/* Return whether the strides of a and b are such that they form an
effective contiguous 1d array. Assumes that a.is >= b.is. */
static int strides_contig(iodim *a, iodim *b)
{
return (a->is == b->is * b->n && a->os == b->os * b->n);
}
/* Like tensor_compress, but also compress into one dimension any
group of dimensions that form a contiguous block of indices with
some stride. (This can safely be done for transform vector sizes.) */
tensor *X(tensor_compress_contiguous)(const tensor *sz)
{
int i, rnk;
tensor *sz2, *x;
if (X(tensor_sz)(sz) == 0)
return X(mktensor)(RNK_MINFTY);
sz2 = X(tensor_compress)(sz);
A(FINITE_RNK(sz2->rnk));
if (sz2->rnk < 2) /* nothing to compress */
return sz2;
for (i = rnk = 1; i < sz2->rnk; ++i)
if (!strides_contig(sz2->dims + i - 1, sz2->dims + i))
++rnk;
x = X(mktensor)(rnk);
x->dims[0] = sz2->dims[0];
for (i = rnk = 1; i < sz2->rnk; ++i) {
if (strides_contig(sz2->dims + i - 1, sz2->dims + i)) {
x->dims[rnk - 1].n *= sz2->dims[i].n;
x->dims[rnk - 1].is = sz2->dims[i].is;
x->dims[rnk - 1].os = sz2->dims[i].os;
} else {
A(rnk < x->rnk);
x->dims[rnk++] = sz2->dims[i];
}
}
X(tensor_destroy)(sz2);
return x;
}
/* The inverse of X(tensor_append): splits the sz tensor into
tensor a followed by tensor b, where a's rank is arnk. */
void X(tensor_split)(const tensor *sz, tensor **a, int arnk, tensor **b)
{
A(FINITE_RNK(sz->rnk) && FINITE_RNK(arnk));
*a = X(tensor_copy_sub)(sz, 0, arnk);
*b = X(tensor_copy_sub)(sz, arnk, sz->rnk - arnk);
}
|
