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
|
/*
* Copyright (c) 2003 Matteo Frigo
* Copyright (c) 2003 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: ct-ditbuf.c,v 1.1 2008/10/17 06:11:08 scuri Exp $ */
/* decimation in time Cooley-Tukey. Codelet operates on
contiguous buffer rather than directly on the output array. */
/* FIXME: find a way to use rank-0 transforms for this stuff */
#include "dft.h"
#include "ct.h"
/*
Copy A -> B, where A and B are n0 x n1 complex matrices
such that the (i0, i1) element has index (i0 * s0 + i1 * s1).
*/
static void cpy(int n0, int n1,
const R *rA, const R *iA, int sa0, int sa1,
R *rB, R *iB, int sb0, int sb1)
{
int i0, i1;
int ima = iA - rA, imb = iB - rB;
for (i0 = 0; i0 < n0; ++i0) {
const R *pa;
R *pb;
pa = rA; rA += sa0;
pb = rB; rB += sb0;
for (i1 = 0; i1 < n1; ++i1) {
R xr = pa[0], xi = pa[ima];
pb[0] = xr; pb[imb] = xi;
pa += sa1; pb += sb1;
}
}
}
static const R *doit(kdft_dit k, R *rA, R *iA, const R *W, int ios, int dist,
int r, int batchsz, R *buf, stride bufstride)
{
cpy(r, batchsz, rA, iA, ios, dist, buf, buf + 1, 2, 2 * r);
W = k(buf, buf + 1, W, bufstride, batchsz, 2 * r);
cpy(r, batchsz, buf, buf + 1, 2, 2 * r, rA, iA, ios, dist);
return W;
}
#define BATCHSZ 4 /* FIXME: parametrize? */
static void apply(const plan *ego_, R *ri, R *ii, R *ro, R *io)
{
const plan_ct *ego = (const plan_ct *) ego_;
plan *cld0 = ego->cld;
plan_dft *cld = (plan_dft *) cld0;
/* two-dimensional r x vl sub-transform: */
cld->apply(cld0, ri, ii, ro, io);
{
int i, j, m = ego->m, vl = ego->vl, r = ego->r;
int os = ego->os, ovs = ego->ovs, ios = ego->iios;
R *buf;
STACK_MALLOC(R *, buf, r * BATCHSZ * 2 * sizeof(R));
for (i = 0; i < vl; ++i) {
R *rA = ro + i * ovs, *iA = io + i * ovs;
const R *W = ego->td->W;
for (j = m; j >= BATCHSZ; j -= BATCHSZ) {
W = doit(ego->k.dit, rA, iA, W, ios, os, r,
BATCHSZ, buf, ego->vs);
rA += os * (int)BATCHSZ;
iA += os * (int)BATCHSZ;
}
/* do remaining j calls, if any */
if (j > 0)
doit(ego->k.dit, rA, iA, W, ios, os, r, j, buf, ego->vs);
}
STACK_FREE(buf);
}
}
static int applicable0(const solver_ct *ego, const problem *p_,
const planner *plnr)
{
UNUSED(plnr);
if (X(dft_ct_applicable)(ego, p_)) {
const ct_desc *e = ego->desc;
const problem_dft *p = (const problem_dft *) p_;
iodim *d = p->sz->dims;
int m = d[0].n / e->radix;
return (1
/* check both batch size and remainder */
&& (m < BATCHSZ ||
(e->genus->okp(e, 0, ((const R *)0)+1, 2, 0, BATCHSZ,
2 * e->radix, plnr)))
&& (e->genus->okp(e, 0, ((const R *)0)+1, 2, 0, m % BATCHSZ,
2 * e->radix, plnr))
);
}
return 0;
}
static int applicable(const solver_ct *ego, const problem *p_,
const planner *plnr)
{
const problem_dft *p;
if (!applicable0(ego, p_, plnr)) return 0;
p = (const problem_dft *) p_;
/* emulate fftw2 behavior */
if (NO_VRECURSEP(plnr) && (p->vecsz->rnk > 0)) return 0;
if (NO_UGLYP(plnr) && X(ct_uglyp)(512, p->sz->dims[0].n,
ego->desc->radix))
return 0;
return 1;
}
static void finish(plan_ct *ego)
{
const ct_desc *d = ego->slv->desc;
ego->iios = ego->m * ego->os;
ego->vs = X(mkstride)(ego->r, 2);
X(ops_madd)(ego->vl * ego->m / d->genus->vl, &d->ops, &ego->cld->ops,
&ego->super.super.ops);
/* 4 load/stores * N * VL */
ego->super.super.ops.other += 4 * ego->r * ego->m * ego->vl;
}
static plan *mkplan(const solver *ego, const problem *p, planner *plnr)
{
static const ctadt adt = {
sizeof(plan_ct), X(dft_mkcld_dit), finish, applicable, apply
};
return X(mkplan_dft_ct)((const solver_ct *) ego, p, plnr, &adt);
}
solver *X(mksolver_dft_ct_ditbuf)(kdft_dit codelet, const ct_desc *desc)
{
static const solver_adt sadt = { mkplan };
static const char name[] = "dft-ditbuf";
union kct k;
k.dit = codelet;
return X(mksolver_dft_ct)(k, desc, name, &sadt);
}
|
