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
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
|
/*
* Copyright (c) 2003, 2007-8 Matteo Frigo
* Copyright (c) 2003, 2007-8 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
*
*/
/* out of place 2D copy routines */
#include "ifftw.h"
#if defined(__x86_64__) || defined(_M_X64) || defined(_M_AMD64)
# ifdef HAVE_XMMINTRIN_H
# include <xmmintrin.h>
# define WIDE_TYPE __m128
# endif
#endif
#ifndef WIDE_TYPE
/* fall back to double, which means that WIDE_TYPE will be unused */
# define WIDE_TYPE double
#endif
void X(cpy2d)(R *I, R *O,
INT n0, INT is0, INT os0,
INT n1, INT is1, INT os1,
INT vl)
{
INT i0, i1, v;
switch (vl) {
case 1:
for (i1 = 0; i1 < n1; ++i1)
for (i0 = 0; i0 < n0; ++i0) {
R x0 = I[i0 * is0 + i1 * is1];
O[i0 * os0 + i1 * os1] = x0;
}
break;
case 2:
if (1
&& (2 * sizeof(R) == sizeof(WIDE_TYPE))
&& (sizeof(WIDE_TYPE) > sizeof(double))
&& (((size_t)I) % sizeof(WIDE_TYPE) == 0)
&& (((size_t)O) % sizeof(WIDE_TYPE) == 0)
&& ((is0 & 1) == 0)
&& ((is1 & 1) == 0)
&& ((os0 & 1) == 0)
&& ((os1 & 1) == 0)) {
/* copy R[2] as WIDE_TYPE if WIDE_TYPE is large
enough to hold R[2], and if the input is
properly aligned. This is a win when R==double
and WIDE_TYPE is 128 bits. */
for (i1 = 0; i1 < n1; ++i1)
for (i0 = 0; i0 < n0; ++i0) {
*(WIDE_TYPE *)&O[i0 * os0 + i1 * os1] =
*(WIDE_TYPE *)&I[i0 * is0 + i1 * is1];
}
} else if (1
&& (2 * sizeof(R) == sizeof(double))
&& (((size_t)I) % sizeof(double) == 0)
&& (((size_t)O) % sizeof(double) == 0)
&& ((is0 & 1) == 0)
&& ((is1 & 1) == 0)
&& ((os0 & 1) == 0)
&& ((os1 & 1) == 0)) {
/* copy R[2] as double if double is large enough to
hold R[2], and if the input is properly aligned.
This case applies when R==float */
for (i1 = 0; i1 < n1; ++i1)
for (i0 = 0; i0 < n0; ++i0) {
*(double *)&O[i0 * os0 + i1 * os1] =
*(double *)&I[i0 * is0 + i1 * is1];
}
} else {
for (i1 = 0; i1 < n1; ++i1)
for (i0 = 0; i0 < n0; ++i0) {
R x0 = I[i0 * is0 + i1 * is1];
R x1 = I[i0 * is0 + i1 * is1 + 1];
O[i0 * os0 + i1 * os1] = x0;
O[i0 * os0 + i1 * os1 + 1] = x1;
}
}
break;
default:
for (i1 = 0; i1 < n1; ++i1)
for (i0 = 0; i0 < n0; ++i0)
for (v = 0; v < vl; ++v) {
R x0 = I[i0 * is0 + i1 * is1 + v];
O[i0 * os0 + i1 * os1 + v] = x0;
}
break;
}
}
/* like cpy2d, but read input contiguously if possible */
void X(cpy2d_ci)(R *I, R *O,
INT n0, INT is0, INT os0,
INT n1, INT is1, INT os1,
INT vl)
{
if (IABS(is0) < IABS(is1)) /* inner loop is for n0 */
X(cpy2d) (I, O, n0, is0, os0, n1, is1, os1, vl);
else
X(cpy2d) (I, O, n1, is1, os1, n0, is0, os0, vl);
}
/* like cpy2d, but write output contiguously if possible */
void X(cpy2d_co)(R *I, R *O,
INT n0, INT is0, INT os0,
INT n1, INT is1, INT os1,
INT vl)
{
if (IABS(os0) < IABS(os1)) /* inner loop is for n0 */
X(cpy2d) (I, O, n0, is0, os0, n1, is1, os1, vl);
else
X(cpy2d) (I, O, n1, is1, os1, n0, is0, os0, vl);
}
/* tiled copy routines */
struct cpy2d_closure {
R *I, *O;
INT is0, os0, is1, os1, vl;
R *buf;
};
static void dotile(INT n0l, INT n0u, INT n1l, INT n1u, void *args)
{
struct cpy2d_closure *k = (struct cpy2d_closure *)args;
X(cpy2d)(k->I + n0l * k->is0 + n1l * k->is1,
k->O + n0l * k->os0 + n1l * k->os1,
n0u - n0l, k->is0, k->os0,
n1u - n1l, k->is1, k->os1,
k->vl);
}
static void dotile_buf(INT n0l, INT n0u, INT n1l, INT n1u, void *args)
{
struct cpy2d_closure *k = (struct cpy2d_closure *)args;
/* copy from I to buf */
X(cpy2d_ci)(k->I + n0l * k->is0 + n1l * k->is1,
k->buf,
n0u - n0l, k->is0, k->vl,
n1u - n1l, k->is1, k->vl * (n0u - n0l),
k->vl);
/* copy from buf to O */
X(cpy2d_co)(k->buf,
k->O + n0l * k->os0 + n1l * k->os1,
n0u - n0l, k->vl, k->os0,
n1u - n1l, k->vl * (n0u - n0l), k->os1,
k->vl);
}
void X(cpy2d_tiled)(R *I, R *O,
INT n0, INT is0, INT os0,
INT n1, INT is1, INT os1, INT vl)
{
INT tilesz = X(compute_tilesz)(vl,
1 /* input array */
+ 1 /* ouput array */);
struct cpy2d_closure k;
k.I = I;
k.O = O;
k.is0 = is0;
k.os0 = os0;
k.is1 = is1;
k.os1 = os1;
k.vl = vl;
k.buf = 0; /* unused */
X(tile2d)(0, n0, 0, n1, tilesz, dotile, &k);
}
void X(cpy2d_tiledbuf)(R *I, R *O,
INT n0, INT is0, INT os0,
INT n1, INT is1, INT os1, INT vl)
{
R buf[CACHESIZE / (2 * sizeof(R))];
/* input and buffer in cache, or
output and buffer in cache */
INT tilesz = X(compute_tilesz)(vl, 2);
struct cpy2d_closure k;
k.I = I;
k.O = O;
k.is0 = is0;
k.os0 = os0;
k.is1 = is1;
k.os1 = os1;
k.vl = vl;
k.buf = buf;
A(tilesz * tilesz * vl * sizeof(R) <= sizeof(buf));
X(tile2d)(0, n0, 0, n1, tilesz, dotile_buf, &k);
}
|
