To check out this repository please hg clone the following URL, or open the URL using EasyMercurial or your preferred Mercurial client.
The primary repository for this project is hosted at https://github.com/sonic-visualiser/sv-dependency-builds .
This repository is a read-only copy which is updated automatically every hour.
root / src / fftw-3.3.8 / dft / dftw-genericbuf.c @ 167:bd3cc4d1df30
History | View | Annotate | Download (5.69 KB)
| 1 |
/*
|
|---|---|
| 2 |
* Copyright (c) 2003, 2007-14 Matteo Frigo
|
| 3 |
* Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
|
| 4 |
*
|
| 5 |
* This program is free software; you can redistribute it and/or modify
|
| 6 |
* it under the terms of the GNU General Public License as published by
|
| 7 |
* the Free Software Foundation; either version 2 of the License, or
|
| 8 |
* (at your option) any later version.
|
| 9 |
*
|
| 10 |
* This program is distributed in the hope that it will be useful,
|
| 11 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 12 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
| 13 |
* GNU General Public License for more details.
|
| 14 |
*
|
| 15 |
* You should have received a copy of the GNU General Public License
|
| 16 |
* along with this program; if not, write to the Free Software
|
| 17 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
| 18 |
*
|
| 19 |
*/
|
| 20 |
|
| 21 |
/* express a twiddle problem in terms of dft + multiplication by
|
| 22 |
twiddle factors */
|
| 23 |
|
| 24 |
#include "dft/ct.h" |
| 25 |
|
| 26 |
typedef struct { |
| 27 |
ct_solver super; |
| 28 |
INT batchsz; |
| 29 |
} S; |
| 30 |
|
| 31 |
typedef struct { |
| 32 |
plan_dftw super; |
| 33 |
|
| 34 |
INT r, rs, m, ms, v, vs, mb, me; |
| 35 |
INT batchsz; |
| 36 |
plan *cld; |
| 37 |
|
| 38 |
triggen *t; |
| 39 |
const S *slv;
|
| 40 |
} P; |
| 41 |
|
| 42 |
|
| 43 |
#define BATCHDIST(r) ((r) + 16) |
| 44 |
|
| 45 |
/**************************************************************/
|
| 46 |
static void bytwiddle(const P *ego, INT mb, INT me, R *buf, R *rio, R *iio) |
| 47 |
{
|
| 48 |
INT j, k; |
| 49 |
INT r = ego->r, rs = ego->rs, ms = ego->ms; |
| 50 |
triggen *t = ego->t; |
| 51 |
for (j = 0; j < r; ++j) { |
| 52 |
for (k = mb; k < me; ++k)
|
| 53 |
t->rotate(t, j * k, |
| 54 |
rio[j * rs + k * ms], |
| 55 |
iio[j * rs + k * ms], |
| 56 |
&buf[j * 2 + 2 * BATCHDIST(r) * (k - mb) + 0]); |
| 57 |
} |
| 58 |
} |
| 59 |
|
| 60 |
static int applicable0(const S *ego, |
| 61 |
INT r, INT irs, INT ors, |
| 62 |
INT m, INT v, |
| 63 |
INT mcount) |
| 64 |
{
|
| 65 |
return (1 |
| 66 |
&& v == 1
|
| 67 |
&& irs == ors |
| 68 |
&& mcount >= ego->batchsz |
| 69 |
&& mcount % ego->batchsz == 0
|
| 70 |
&& r >= 64
|
| 71 |
&& m >= r |
| 72 |
); |
| 73 |
} |
| 74 |
|
| 75 |
static int applicable(const S *ego, |
| 76 |
INT r, INT irs, INT ors, |
| 77 |
INT m, INT v, |
| 78 |
INT mcount, |
| 79 |
const planner *plnr)
|
| 80 |
{
|
| 81 |
if (!applicable0(ego, r, irs, ors, m, v, mcount))
|
| 82 |
return 0; |
| 83 |
if (NO_UGLYP(plnr) && m * r < 65536) |
| 84 |
return 0; |
| 85 |
|
| 86 |
return 1; |
| 87 |
} |
| 88 |
|
| 89 |
static void dobatch(const P *ego, INT mb, INT me, R *buf, R *rio, R *iio) |
| 90 |
{
|
| 91 |
plan_dft *cld; |
| 92 |
INT ms = ego->ms; |
| 93 |
|
| 94 |
bytwiddle(ego, mb, me, buf, rio, iio); |
| 95 |
|
| 96 |
cld = (plan_dft *) ego->cld; |
| 97 |
cld->apply(ego->cld, buf, buf + 1, buf, buf + 1); |
| 98 |
X(cpy2d_pair_co)(buf, buf + 1,
|
| 99 |
rio + ms * mb, iio + ms * mb, |
| 100 |
me-mb, 2 * BATCHDIST(ego->r), ms,
|
| 101 |
ego->r, 2, ego->rs);
|
| 102 |
} |
| 103 |
|
| 104 |
static void apply(const plan *ego_, R *rio, R *iio) |
| 105 |
{
|
| 106 |
const P *ego = (const P *) ego_; |
| 107 |
R *buf = (R *) MALLOC(sizeof(R) * 2 * BATCHDIST(ego->r) * ego->batchsz, |
| 108 |
BUFFERS); |
| 109 |
INT m; |
| 110 |
|
| 111 |
for (m = ego->mb; m < ego->me; m += ego->batchsz)
|
| 112 |
dobatch(ego, m, m + ego->batchsz, buf, rio, iio); |
| 113 |
|
| 114 |
A(m == ego->me); |
| 115 |
|
| 116 |
X(ifree)(buf); |
| 117 |
} |
| 118 |
|
| 119 |
static void awake(plan *ego_, enum wakefulness wakefulness) |
| 120 |
{
|
| 121 |
P *ego = (P *) ego_; |
| 122 |
X(plan_awake)(ego->cld, wakefulness); |
| 123 |
|
| 124 |
switch (wakefulness) {
|
| 125 |
case SLEEPY:
|
| 126 |
X(triggen_destroy)(ego->t); ego->t = 0;
|
| 127 |
break;
|
| 128 |
default:
|
| 129 |
ego->t = X(mktriggen)(AWAKE_SQRTN_TABLE, ego->r * ego->m); |
| 130 |
break;
|
| 131 |
} |
| 132 |
} |
| 133 |
|
| 134 |
static void destroy(plan *ego_) |
| 135 |
{
|
| 136 |
P *ego = (P *) ego_; |
| 137 |
X(plan_destroy_internal)(ego->cld); |
| 138 |
} |
| 139 |
|
| 140 |
static void print(const plan *ego_, printer *p) |
| 141 |
{
|
| 142 |
const P *ego = (const P *) ego_; |
| 143 |
p->print(p, "(dftw-genericbuf/%D-%D-%D%(%p%))",
|
| 144 |
ego->batchsz, ego->r, ego->m, ego->cld); |
| 145 |
} |
| 146 |
|
| 147 |
static plan *mkcldw(const ct_solver *ego_, |
| 148 |
INT r, INT irs, INT ors, |
| 149 |
INT m, INT ms, |
| 150 |
INT v, INT ivs, INT ovs, |
| 151 |
INT mstart, INT mcount, |
| 152 |
R *rio, R *iio, |
| 153 |
planner *plnr) |
| 154 |
{
|
| 155 |
const S *ego = (const S *)ego_; |
| 156 |
P *pln; |
| 157 |
plan *cld = 0;
|
| 158 |
R *buf; |
| 159 |
|
| 160 |
static const plan_adt padt = { |
| 161 |
0, awake, print, destroy
|
| 162 |
}; |
| 163 |
|
| 164 |
UNUSED(ivs); UNUSED(ovs); UNUSED(rio); UNUSED(iio); |
| 165 |
|
| 166 |
A(mstart >= 0 && mstart + mcount <= m);
|
| 167 |
if (!applicable(ego, r, irs, ors, m, v, mcount, plnr))
|
| 168 |
return (plan *)0; |
| 169 |
|
| 170 |
buf = (R *) MALLOC(sizeof(R) * 2 * BATCHDIST(r) * ego->batchsz, BUFFERS); |
| 171 |
cld = X(mkplan_d)(plnr, |
| 172 |
X(mkproblem_dft_d)( |
| 173 |
X(mktensor_1d)(r, 2, 2), |
| 174 |
X(mktensor_1d)(ego->batchsz, |
| 175 |
2 * BATCHDIST(r),
|
| 176 |
2 * BATCHDIST(r)),
|
| 177 |
buf, buf + 1, buf, buf + 1 |
| 178 |
) |
| 179 |
); |
| 180 |
X(ifree)(buf); |
| 181 |
if (!cld) goto nada; |
| 182 |
|
| 183 |
pln = MKPLAN_DFTW(P, &padt, apply); |
| 184 |
pln->slv = ego; |
| 185 |
pln->cld = cld; |
| 186 |
pln->r = r; |
| 187 |
pln->m = m; |
| 188 |
pln->ms = ms; |
| 189 |
pln->rs = irs; |
| 190 |
pln->batchsz = ego->batchsz; |
| 191 |
pln->mb = mstart; |
| 192 |
pln->me = mstart + mcount; |
| 193 |
|
| 194 |
{
|
| 195 |
double n0 = (r - 1) * (mcount - 1); |
| 196 |
pln->super.super.ops = cld->ops; |
| 197 |
pln->super.super.ops.mul += 8 * n0;
|
| 198 |
pln->super.super.ops.add += 4 * n0;
|
| 199 |
pln->super.super.ops.other += 8 * n0;
|
| 200 |
} |
| 201 |
return &(pln->super.super);
|
| 202 |
|
| 203 |
nada:
|
| 204 |
X(plan_destroy_internal)(cld); |
| 205 |
return (plan *) 0; |
| 206 |
} |
| 207 |
|
| 208 |
static void regsolver(planner *plnr, INT r, INT batchsz) |
| 209 |
{
|
| 210 |
S *slv = (S *)X(mksolver_ct)(sizeof(S), r, DECDIT, mkcldw, 0); |
| 211 |
slv->batchsz = batchsz; |
| 212 |
REGISTER_SOLVER(plnr, &(slv->super.super)); |
| 213 |
|
| 214 |
if (X(mksolver_ct_hook)) {
|
| 215 |
slv = (S *)X(mksolver_ct_hook)(sizeof(S), r, DECDIT, mkcldw, 0); |
| 216 |
slv->batchsz = batchsz; |
| 217 |
REGISTER_SOLVER(plnr, &(slv->super.super)); |
| 218 |
} |
| 219 |
|
| 220 |
} |
| 221 |
|
| 222 |
void X(ct_genericbuf_register)(planner *p)
|
| 223 |
{
|
| 224 |
static const INT radices[] = { -1, -2, -4, -8, -16, -32, -64 }; |
| 225 |
static const INT batchsizes[] = { 4, 8, 16, 32, 64 }; |
| 226 |
unsigned i, j;
|
| 227 |
|
| 228 |
for (i = 0; i < sizeof(radices) / sizeof(radices[0]); ++i) |
| 229 |
for (j = 0; j < sizeof(batchsizes) / sizeof(batchsizes[0]); ++j) |
| 230 |
regsolver(p, radices[i], batchsizes[j]); |
| 231 |
} |