static plan *mkplan(const solver *ego_, const problem *p_, planner *plnr)
{
const S *ego = (const S *) ego_;
P *pln;
const problem_dft *p;
iodim *d;
const kdft_desc *e = ego->desc;
static const plan_adt padt = {
X(dft_solve), X(null_awake), print, destroy
};
UNUSED(plnr);
if (ego->bufferedp) {
if (!applicable_buf(ego_, p_, plnr))
return (plan *)0;
pln = MKPLAN_DFT(P, &padt, apply_buf);
} else {
int extra_iterp = 0;
if (!applicable(ego_, p_, plnr, &extra_iterp))
return (plan *)0;
pln = MKPLAN_DFT(P, &padt, extra_iterp ? apply_extra_iter : apply);
}
p = (const problem_dft *) p_;
d = p->sz->dims;
pln->k = ego->k;
pln->n = d[0].n;
pln->is = X(mkstride)(pln->n, d[0].is);
pln->os = X(mkstride)(pln->n, d[0].os);
pln->bufstride = X(mkstride)(pln->n, 2 * compute_batchsize(pln->n));
X(tensor_tornk1)(p->vecsz, &pln->vl, &pln->ivs, &pln->ovs);
pln->slv = ego;
X(ops_zero)(&pln->super.super.ops);
X(ops_madd2)(pln->vl / e->genus->vl, &e->ops, &pln->super.super.ops);
if (ego->bufferedp)
pln->super.super.ops.other += 4 * pln->n * pln->vl;
pln->super.super.could_prune_now_p = !ego->bufferedp;
return &(pln->super.super);
}
static plan *mkplan(const solver *ego_, const problem *p_, planner *plnr)
{
const S *ego = (const S *) ego_;
P *pln;
const problem_rdft *p;
iodim *d;
INT rs, cs, b, n;
static const plan_adt padt = {
X(rdft_solve), X(null_awake), print, destroy
};
UNUSED(plnr);
if (ego->bufferedp) {
if (!applicable_buf(ego_, p_))
return (plan *)0;
} else {
if (!applicable(ego_, p_))
return (plan *)0;
}
p = (const problem_rdft *) p_;
if (R2HC_KINDP(p->kind[0])) {
rs = p->sz->dims[0].is; cs = p->sz->dims[0].os;
pln = MKPLAN_RDFT(P, &padt,
ego->bufferedp ? apply_buf_r2hc : apply_r2hc);
} else {
rs = p->sz->dims[0].os; cs = p->sz->dims[0].is;
pln = MKPLAN_RDFT(P, &padt,
ego->bufferedp ? apply_buf_hc2r : apply_hc2r);
}
d = p->sz->dims;
n = d[0].n;
pln->k = ego->k;
pln->n = n;
pln->rs0 = rs;
pln->rs = X(mkstride)(n, 2 * rs);
pln->csr = X(mkstride)(n, cs);
pln->csi = X(mkstride)(n, -cs);
pln->ioffset = ioffset(p->kind[0], n, cs);
b = compute_batchsize(n);
pln->brs = X(mkstride)(n, 2 * b);
pln->bcsr = X(mkstride)(n, b);
pln->bcsi = X(mkstride)(n, -b);
pln->bioffset = ioffset(p->kind[0], n, b);
X(tensor_tornk1)(p->vecsz, &pln->vl, &pln->ivs, &pln->ovs);
pln->slv = ego;
X(ops_zero)(&pln->super.super.ops);
X(ops_madd2)(pln->vl / ego->desc->genus->vl,
&ego->desc->ops,
&pln->super.super.ops);
if (ego->bufferedp)
pln->super.super.ops.other += 2 * n * pln->vl;
pln->super.super.could_prune_now_p = !ego->bufferedp;
return &(pln->super.super);
}
