232 lines
5.7 KiB
C
232 lines
5.7 KiB
C
/*
|
|
* Copyright (c) 2003, 2007-14 Matteo Frigo
|
|
* Copyright (c) 2003, 2007-14 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
*
|
|
*/
|
|
|
|
/* express a twiddle problem in terms of dft + multiplication by
|
|
twiddle factors */
|
|
|
|
#include "dft/ct.h"
|
|
|
|
typedef struct {
|
|
ct_solver super;
|
|
INT batchsz;
|
|
} S;
|
|
|
|
typedef struct {
|
|
plan_dftw super;
|
|
|
|
INT r, rs, m, ms, v, vs, mb, me;
|
|
INT batchsz;
|
|
plan *cld;
|
|
|
|
triggen *t;
|
|
const S *slv;
|
|
} P;
|
|
|
|
|
|
#define BATCHDIST(r) ((r) + 16)
|
|
|
|
/**************************************************************/
|
|
static void bytwiddle(const P *ego, INT mb, INT me, R *buf, R *rio, R *iio)
|
|
{
|
|
INT j, k;
|
|
INT r = ego->r, rs = ego->rs, ms = ego->ms;
|
|
triggen *t = ego->t;
|
|
for (j = 0; j < r; ++j) {
|
|
for (k = mb; k < me; ++k)
|
|
t->rotate(t, j * k,
|
|
rio[j * rs + k * ms],
|
|
iio[j * rs + k * ms],
|
|
&buf[j * 2 + 2 * BATCHDIST(r) * (k - mb) + 0]);
|
|
}
|
|
}
|
|
|
|
static int applicable0(const S *ego,
|
|
INT r, INT irs, INT ors,
|
|
INT m, INT v,
|
|
INT mcount)
|
|
{
|
|
return (1
|
|
&& v == 1
|
|
&& irs == ors
|
|
&& mcount >= ego->batchsz
|
|
&& mcount % ego->batchsz == 0
|
|
&& r >= 64
|
|
&& m >= r
|
|
);
|
|
}
|
|
|
|
static int applicable(const S *ego,
|
|
INT r, INT irs, INT ors,
|
|
INT m, INT v,
|
|
INT mcount,
|
|
const planner *plnr)
|
|
{
|
|
if (!applicable0(ego, r, irs, ors, m, v, mcount))
|
|
return 0;
|
|
if (NO_UGLYP(plnr) && m * r < 65536)
|
|
return 0;
|
|
|
|
return 1;
|
|
}
|
|
|
|
static void dobatch(const P *ego, INT mb, INT me, R *buf, R *rio, R *iio)
|
|
{
|
|
plan_dft *cld;
|
|
INT ms = ego->ms;
|
|
|
|
bytwiddle(ego, mb, me, buf, rio, iio);
|
|
|
|
cld = (plan_dft *) ego->cld;
|
|
cld->apply(ego->cld, buf, buf + 1, buf, buf + 1);
|
|
X(cpy2d_pair_co)(buf, buf + 1,
|
|
rio + ms * mb, iio + ms * mb,
|
|
me-mb, 2 * BATCHDIST(ego->r), ms,
|
|
ego->r, 2, ego->rs);
|
|
}
|
|
|
|
static void apply(const plan *ego_, R *rio, R *iio)
|
|
{
|
|
const P *ego = (const P *) ego_;
|
|
R *buf = (R *) MALLOC(sizeof(R) * 2 * BATCHDIST(ego->r) * ego->batchsz,
|
|
BUFFERS);
|
|
INT m;
|
|
|
|
for (m = ego->mb; m < ego->me; m += ego->batchsz)
|
|
dobatch(ego, m, m + ego->batchsz, buf, rio, iio);
|
|
|
|
A(m == ego->me);
|
|
|
|
X(ifree)(buf);
|
|
}
|
|
|
|
static void awake(plan *ego_, enum wakefulness wakefulness)
|
|
{
|
|
P *ego = (P *) ego_;
|
|
X(plan_awake)(ego->cld, wakefulness);
|
|
|
|
switch (wakefulness) {
|
|
case SLEEPY:
|
|
X(triggen_destroy)(ego->t); ego->t = 0;
|
|
break;
|
|
default:
|
|
ego->t = X(mktriggen)(AWAKE_SQRTN_TABLE, ego->r * ego->m);
|
|
break;
|
|
}
|
|
}
|
|
|
|
static void destroy(plan *ego_)
|
|
{
|
|
P *ego = (P *) ego_;
|
|
X(plan_destroy_internal)(ego->cld);
|
|
}
|
|
|
|
static void print(const plan *ego_, printer *p)
|
|
{
|
|
const P *ego = (const P *) ego_;
|
|
p->print(p, "(dftw-genericbuf/%D-%D-%D%(%p%))",
|
|
ego->batchsz, ego->r, ego->m, ego->cld);
|
|
}
|
|
|
|
static plan *mkcldw(const ct_solver *ego_,
|
|
INT r, INT irs, INT ors,
|
|
INT m, INT ms,
|
|
INT v, INT ivs, INT ovs,
|
|
INT mstart, INT mcount,
|
|
R *rio, R *iio,
|
|
planner *plnr)
|
|
{
|
|
const S *ego = (const S *)ego_;
|
|
P *pln;
|
|
plan *cld = 0;
|
|
R *buf;
|
|
|
|
static const plan_adt padt = {
|
|
0, awake, print, destroy
|
|
};
|
|
|
|
UNUSED(ivs); UNUSED(ovs); UNUSED(rio); UNUSED(iio);
|
|
|
|
A(mstart >= 0 && mstart + mcount <= m);
|
|
if (!applicable(ego, r, irs, ors, m, v, mcount, plnr))
|
|
return (plan *)0;
|
|
|
|
buf = (R *) MALLOC(sizeof(R) * 2 * BATCHDIST(r) * ego->batchsz, BUFFERS);
|
|
cld = X(mkplan_d)(plnr,
|
|
X(mkproblem_dft_d)(
|
|
X(mktensor_1d)(r, 2, 2),
|
|
X(mktensor_1d)(ego->batchsz,
|
|
2 * BATCHDIST(r),
|
|
2 * BATCHDIST(r)),
|
|
buf, buf + 1, buf, buf + 1
|
|
)
|
|
);
|
|
X(ifree)(buf);
|
|
if (!cld) goto nada;
|
|
|
|
pln = MKPLAN_DFTW(P, &padt, apply);
|
|
pln->slv = ego;
|
|
pln->cld = cld;
|
|
pln->r = r;
|
|
pln->m = m;
|
|
pln->ms = ms;
|
|
pln->rs = irs;
|
|
pln->batchsz = ego->batchsz;
|
|
pln->mb = mstart;
|
|
pln->me = mstart + mcount;
|
|
|
|
{
|
|
double n0 = (r - 1) * (mcount - 1);
|
|
pln->super.super.ops = cld->ops;
|
|
pln->super.super.ops.mul += 8 * n0;
|
|
pln->super.super.ops.add += 4 * n0;
|
|
pln->super.super.ops.other += 8 * n0;
|
|
}
|
|
return &(pln->super.super);
|
|
|
|
nada:
|
|
X(plan_destroy_internal)(cld);
|
|
return (plan *) 0;
|
|
}
|
|
|
|
static void regsolver(planner *plnr, INT r, INT batchsz)
|
|
{
|
|
S *slv = (S *)X(mksolver_ct)(sizeof(S), r, DECDIT, mkcldw, 0);
|
|
slv->batchsz = batchsz;
|
|
REGISTER_SOLVER(plnr, &(slv->super.super));
|
|
|
|
if (X(mksolver_ct_hook)) {
|
|
slv = (S *)X(mksolver_ct_hook)(sizeof(S), r, DECDIT, mkcldw, 0);
|
|
slv->batchsz = batchsz;
|
|
REGISTER_SOLVER(plnr, &(slv->super.super));
|
|
}
|
|
|
|
}
|
|
|
|
void X(ct_genericbuf_register)(planner *p)
|
|
{
|
|
static const INT radices[] = { -1, -2, -4, -8, -16, -32, -64 };
|
|
static const INT batchsizes[] = { 4, 8, 16, 32, 64 };
|
|
unsigned i, j;
|
|
|
|
for (i = 0; i < sizeof(radices) / sizeof(radices[0]); ++i)
|
|
for (j = 0; j < sizeof(batchsizes) / sizeof(batchsizes[0]); ++j)
|
|
regsolver(p, radices[i], batchsizes[j]);
|
|
}
|