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
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
|
/*
* 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
*
*/
#include "ifftw.h"
/* in place square transposition, iterative */
void X(transpose)(R *I, INT n, INT s0, INT s1, INT vl)
{
INT i0, i1, v;
switch (vl) {
case 1:
for (i1 = 1; i1 < n; ++i1) {
for (i0 = 0; i0 < i1; ++i0) {
R x0 = I[i1 * s0 + i0 * s1];
R y0 = I[i1 * s1 + i0 * s0];
I[i1 * s1 + i0 * s0] = x0;
I[i1 * s0 + i0 * s1] = y0;
}
}
break;
case 2:
for (i1 = 1; i1 < n; ++i1) {
for (i0 = 0; i0 < i1; ++i0) {
R x0 = I[i1 * s0 + i0 * s1];
R x1 = I[i1 * s0 + i0 * s1 + 1];
R y0 = I[i1 * s1 + i0 * s0];
R y1 = I[i1 * s1 + i0 * s0 + 1];
I[i1 * s1 + i0 * s0] = x0;
I[i1 * s1 + i0 * s0 + 1] = x1;
I[i1 * s0 + i0 * s1] = y0;
I[i1 * s0 + i0 * s1 + 1] = y1;
}
}
break;
default:
for (i1 = 1; i1 < n; ++i1) {
for (i0 = 0; i0 < i1; ++i0) {
for (v = 0; v < vl; ++v) {
R x0 = I[i1 * s0 + i0 * s1 + v];
R y0 = I[i1 * s1 + i0 * s0 + v];
I[i1 * s1 + i0 * s0 + v] = x0;
I[i1 * s0 + i0 * s1 + v] = y0;
}
}
}
break;
}
}
struct transpose_closure {
R *I;
INT s0, s1, vl, tilesz;
R *buf0, *buf1;
};
static void dotile(INT n0l, INT n0u, INT n1l, INT n1u, void *args)
{
struct transpose_closure *k = (struct transpose_closure *)args;
R *I = k->I;
INT s0 = k->s0, s1 = k->s1, vl = k->vl;
INT i0, i1, v;
switch (vl) {
case 1:
for (i1 = n1l; i1 < n1u; ++i1) {
for (i0 = n0l; i0 < n0u; ++i0) {
R x0 = I[i1 * s0 + i0 * s1];
R y0 = I[i1 * s1 + i0 * s0];
I[i1 * s1 + i0 * s0] = x0;
I[i1 * s0 + i0 * s1] = y0;
}
}
break;
case 2:
for (i1 = n1l; i1 < n1u; ++i1) {
for (i0 = n0l; i0 < n0u; ++i0) {
R x0 = I[i1 * s0 + i0 * s1];
R x1 = I[i1 * s0 + i0 * s1 + 1];
R y0 = I[i1 * s1 + i0 * s0];
R y1 = I[i1 * s1 + i0 * s0 + 1];
I[i1 * s1 + i0 * s0] = x0;
I[i1 * s1 + i0 * s0 + 1] = x1;
I[i1 * s0 + i0 * s1] = y0;
I[i1 * s0 + i0 * s1 + 1] = y1;
}
}
break;
default:
for (i1 = n1l; i1 < n1u; ++i1) {
for (i0 = n0l; i0 < n0u; ++i0) {
for (v = 0; v < vl; ++v) {
R x0 = I[i1 * s0 + i0 * s1 + v];
R y0 = I[i1 * s1 + i0 * s0 + v];
I[i1 * s1 + i0 * s0 + v] = x0;
I[i1 * s0 + i0 * s1 + v] = y0;
}
}
}
}
}
static void dotile_buf(INT n0l, INT n0u, INT n1l, INT n1u, void *args)
{
struct transpose_closure *k = (struct transpose_closure *)args;
X(cpy2d_ci)(k->I + n0l * k->s0 + n1l * k->s1,
k->buf0,
n0u - n0l, k->s0, k->vl,
n1u - n1l, k->s1, k->vl * (n0u - n0l),
k->vl);
X(cpy2d_ci)(k->I + n0l * k->s1 + n1l * k->s0,
k->buf1,
n0u - n0l, k->s1, k->vl,
n1u - n1l, k->s0, k->vl * (n0u - n0l),
k->vl);
X(cpy2d_co)(k->buf1,
k->I + n0l * k->s0 + n1l * k->s1,
n0u - n0l, k->vl, k->s0,
n1u - n1l, k->vl * (n0u - n0l), k->s1,
k->vl);
X(cpy2d_co)(k->buf0,
k->I + n0l * k->s1 + n1l * k->s0,
n0u - n0l, k->vl, k->s1,
n1u - n1l, k->vl * (n0u - n0l), k->s0,
k->vl);
}
static void transpose_rec(R *I, INT n,
void (*f)(INT n0l, INT n0u, INT n1l, INT n1u,
void *args),
struct transpose_closure *k)
{
tail:
if (n > 1) {
INT n2 = n / 2;
k->I = I;
X(tile2d)(0, n2, n2, n, k->tilesz, f, k);
transpose_rec(I, n2, f, k);
I += n2 * (k->s0 + k->s1); n -= n2; goto tail;
}
}
void X(transpose_tiled)(R *I, INT n, INT s0, INT s1, INT vl)
{
struct transpose_closure k;
k.s0 = s0;
k.s1 = s1;
k.vl = vl;
/* two blocks must be in cache, to be swapped */
k.tilesz = X(compute_tilesz)(vl, 2);
k.buf0 = k.buf1 = 0; /* unused */
transpose_rec(I, n, dotile, &k);
}
void X(transpose_tiledbuf)(R *I, INT n, INT s0, INT s1, INT vl)
{
struct transpose_closure k;
/* Assume that the the rows of I conflict into the same cache
lines, and therefore we don't need to reserve cache space for
the input. If the rows don't conflict, there is no reason
to use tiledbuf at all.*/
R buf0[CACHESIZE / (2 * sizeof(R))];
R buf1[CACHESIZE / (2 * sizeof(R))];
k.s0 = s0;
k.s1 = s1;
k.vl = vl;
k.tilesz = X(compute_tilesz)(vl, 2);
k.buf0 = buf0;
k.buf1 = buf1;
A(k.tilesz * k.tilesz * vl * sizeof(R) <= sizeof(buf0));
A(k.tilesz * k.tilesz * vl * sizeof(R) <= sizeof(buf1));
transpose_rec(I, n, dotile_buf, &k);
}
|
