/*
* 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
*
*/
#include "rdft.h"
typedef struct {
solver super;
rdft_kind kind;
} S;
typedef struct {
plan_rdft super;
plan *cld;
twid *td;
int os;
int r, m;
rdft_kind kind;
} P;
/***************************************************************************/
static void apply_dit(const plan *ego_, R *I, R *O)
{
const P *ego = (const P *) ego_;
int n, m, r;
int i, j, k;
int os, osm;
E *buf;
const R *W;
R *X, *YO, *YI;
E rsum, isum;
int wp, wincr;
{
plan_rdft *cld = (plan_rdft *) ego->cld;
cld->apply((plan *) cld, I, O);
}
r = ego->r;
STACK_MALLOC(E *, buf, r * 2 * sizeof(E));
osm = (m = ego->m) * (os = ego->os);
n = m * r;
W = ego->td->W;
X = O;
YO = O + r * osm;
YI = O + osm;
/* compute the transform of the r 0th elements (which are real) */
for (i = 0; i + i < r; ++i) {
rsum = K(0.0);
isum = K(0.0);
wincr = m * i;
for (j = 0, wp = 0; j < r; ++j) {
E tw_r = W[2*wp];
E tw_i = W[2*wp+1] ;
E re = X[j * osm];
rsum += re * tw_r;
isum += re * tw_i;
wp += wincr;
if (wp >= n)
wp -= n;
}
buf[2*i] = rsum;
buf[2*i+1] = isum;
}
/* store the transform back onto the A array */
X[0] = buf[0];
for (i = 1; i + i < r; ++i) {
X[i * osm] = buf[2*i];
YO[-i * osm] = buf[2*i+1];
}
X += os;
YI -= os;
YO -= os;
/* compute the transform of the middle elements (which are complex) */
for (k = 1; k + k < m; ++k, X += os, YI -= os, YO -= os) {
for (i = 0; i < r; ++i) {
rsum = K(0.0);
isum = K(0.0);
wincr = k + m * i;
for (j = 0, wp = 0; j < r; ++j) {
E tw_r = W[2*wp];
E tw_i = W[2*wp+1] ;
E re = X[j * osm];
E im = YI[j * osm];
rsum += re * tw_r - im * tw_i;
isum += re * tw_i + im * tw_r;
wp += wincr;
if (wp >= n)
wp -= n;
}
buf[2*i] = rsum;
buf[2*i+1] = isum;
}
/* store the transform back onto the A array */
for (i = 0; i + i < r; ++i) {
X[i * osm] = buf[2*i];
YO[-i * osm] = buf[2*i+1];
}
for (; i < r; ++i) {
X[i * osm] = -buf[2*i+1];
YO[-i * osm] = buf[2*i];
}
}
/* no final element, since m is odd */
STACK_FREE(buf);
}
static void apply_dif(const plan *ego_, R *I, R *O)
{
const P *ego = (const P *) ego_;
int n, m, r;
int i, j, k;
int is, ism;
E *buf;
const R *W;
R *X, *YO, *YI;
E rsum, isum;
int wp, wincr;
r = ego->r;
STACK_MALLOC(E *, buf, r * 2 * sizeof(E));
ism = (m = ego->m) * (is = ego->os);
n = m * r;
W = ego->td->W;
X = I;
YI = I + r * ism;
YO = I + ism;
/*
* compute the transform of the r 0th elements (which are halfcomplex)
* yielding real numbers
*/
/* copy the input into the temporary array */
buf[0] = X[0];
for (i = 1; i + i < r; ++i) {
buf[2*i] = X[i * ism];
buf[2*i+1] = YI[-i * ism];
}
for (i = 0; i < r; ++i) {
rsum = K(0.0);
wincr = m * i;
for (j = 1, wp = wincr; j + j < r; ++j) {
E tw_r = W[2*wp];
E tw_i = W[2*wp+1];
E re = buf[2*j];
E im = buf[2*j+1];
rsum += re * tw_r + im * tw_i;
wp += wincr;
if (wp >= n)
wp -= n;
}
X[i * ism] = K(2.0) * rsum + buf[0];
}
X += is;
YI -= is;
YO -= is;
/* compute the transform of the middle elements (which are complex) */
for (k = 1; k + k < m; ++k, X += is, YI -= is, YO -= is) {
/* copy the input into the temporary array */
for (i = 0; i + i < r; ++i) {
buf[2*i] = X[i * ism];
buf[2*i+1] = YI[-i * ism];
}
for (; i < r; ++i) {
buf[2*i+1] = -X[i * ism];
buf[2*i] = YI[-i * ism];
}
for (i = 0; i < r; ++i) {
rsum = K(0.0);
isum = K(0.0);
wincr = m * i;
for (j = 0, wp = k * i; j < r; ++j) {
E tw_r = W[2*wp];
E tw_i = W[2*wp+1];
E re = buf[2*j];
E im = buf[2*j+1];
rsum += re * tw_r + im * tw_i;
isum += im * tw_r - re * tw_i;
wp += wincr;
if (wp >= n)
wp -= n;
}
X[i * ism] = rsum;
YO[i * ism] = isum;
}
}
/* no final element, since m is odd */
STACK_FREE(buf);
{
plan_rdft *cld = (plan_rdft *) ego->cld;
cld->apply((plan *) cld, I, O);
}
}
/***************************************************************************/
static void awake(plan *ego_, int flg)
{
P *ego = (P *) ego_;
static const tw_instr generic_tw[] = {
{ TW_GENERIC, 0, 0 },
{ TW_NEXT, 1, 0 }
};
AWAKE(ego->cld, flg);
/* FIXME: can we get away with fewer twiddles? */
X(twiddle_awake)(flg, &ego->td, generic_tw,
ego->r * ego->m, ego->r, ego->m);
}
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, "(rdft-generic-%s-%d%(%p%))",
ego->kind == R2HC ? "r2hc-dit" : "hc2r-dif",
ego->r, ego->cld);
}
static int applicable0(const solver *ego_, const problem *p_)
{
if (RDFTP(p_)) {
const S *ego = (const S *) ego_;
const problem_rdft *p = (const problem_rdft *) p_;
return (1
&& p->sz->rnk == 1
&& p->vecsz->rnk == 0
&& p->sz->dims[0].n > 1
&& p->sz->dims[0].n % 2 /* ensure r and n/r odd */
&& p->kind[0] == ego->kind
);
}
return 0;
}
static int applicable(const solver *ego_, const problem *p_,
const planner *plnr)
{
if (NO_UGLYP(plnr)) return 0; /* always ugly */
if (!applicable0(ego_, p_)) return 0;
if (NO_LARGE_GENERICP(plnr)) {
const problem_rdft *p = (const problem_rdft *) p_;
if (X(first_divisor)(p->sz->dims[0].n) >= GENERIC_MIN_BAD) return 0;
}
return 1;
}
static plan *mkplan(const solver *ego, const problem *p_, planner *plnr)
{
const problem_rdft *p = (const problem_rdft *) p_;
P *pln = 0;
int n, r, m;
int is, os;
plan *cld = (plan *) 0;
problem *cldp;
static const plan_adt padt = {
X(rdft_solve), awake, print, destroy
};
if (!applicable(ego, p_, plnr))
goto nada;
n = p->sz->dims[0].n;
is = p->sz->dims[0].is;
os = p->sz->dims[0].os;
r = X(first_divisor)(n);
m = n / r;
if (R2HC_KINDP(p->kind[0])) {
cldp = X(mkproblem_rdft_d)(X(mktensor_1d)(m, r * is, os),
X(mktensor_1d)(r, is, m * os),
p->I, p->O, p->kind);
}
else {
cldp = X(mkproblem_rdft_d)(X(mktensor_1d)(m, is, r * os),
X(mktensor_1d)(r, m * is, os),
p->I, p->O, p->kind);
}
if (!(cld = X(mkplan_d)(plnr, cldp))) goto nada;
pln = MKPLAN_RDFT(P, &padt, R2HC_KINDP(p->kind[0]) ? apply_dit:apply_dif);
pln->os = R2HC_KINDP(p->kind[0]) ? os : is;
pln->r = r;
pln->m = m;
pln->cld = cld;
pln->td = 0;
pln->kind = p->kind[0];
X(ops_zero)(&pln->super.super.ops);
pln->super.super.ops.add = 4 * r * r;
pln->super.super.ops.mul = 4 * r * r;
/* loads + stores, minus loads + stores for all DIT codelets */
pln->super.super.ops.other = 4 * r + 4 * r * r - (6*r - 2);
X(ops_madd)((m - 1)/2, &pln->super.super.ops, &cld->ops,
&pln->super.super.ops);
pln->super.super.ops.add += 2 * r * r;
pln->super.super.ops.mul += 2 * r * r;
pln->super.super.ops.other += 3 * r + 3 * r * r - 2*r;
return &(pln->super.super);
nada:
X(plan_destroy_internal)(cld);
X(ifree0)(pln);
return (plan *) 0;
}
/* constructors */
static solver *mksolver(rdft_kind kind)
{
static const solver_adt sadt = { mkplan };
S *slv = MKSOLVER(S, &sadt);
slv->kind = kind;
return &(slv->super);
}
void X(rdft_generic_register)(planner *p)
{
REGISTER_SOLVER(p, mksolver(R2HC));
REGISTER_SOLVER(p, mksolver(HC2R));
}