/* * 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 * */ #include "dft.h" typedef struct { solver super; } S; typedef struct { plan_dft super; INT n; /* problem size */ INT nb; /* size of convolution */ R *w; /* lambda k . exp(2*pi*i*k^2/(2*n)) */ R *W; /* DFT(w) */ plan *cldf; INT is, os; } P; static void bluestein_sequence(enum wakefulness wakefulness, INT n, R *w) { INT k, ksq, n2 = 2 * n; triggen *t = X(mktriggen)(wakefulness, n2); ksq = 0; for (k = 0; k < n; ++k) { t->cexp(t, ksq, w+2*k); /* careful with overflow */ ksq += 2*k + 1; while (ksq > n2) ksq -= n2; } X(triggen_destroy)(t); } static void mktwiddle(enum wakefulness wakefulness, P *p) { INT i; INT n = p->n, nb = p->nb; R *w, *W; E nbf = (E)nb; p->w = w = (R *) MALLOC(2 * n * sizeof(R), TWIDDLES); p->W = W = (R *) MALLOC(2 * nb * sizeof(R), TWIDDLES); bluestein_sequence(wakefulness, n, w); for (i = 0; i < nb; ++i) W[2*i] = W[2*i+1] = K(0.0); W[0] = w[0] / nbf; W[1] = w[1] / nbf; for (i = 1; i < n; ++i) { W[2*i] = W[2*(nb-i)] = w[2*i] / nbf; W[2*i+1] = W[2*(nb-i)+1] = w[2*i+1] / nbf; } { plan_dft *cldf = (plan_dft *)p->cldf; /* cldf must be awake */ cldf->apply(p->cldf, W, W+1, W, W+1); } } static void apply(const plan *ego_, R *ri, R *ii, R *ro, R *io) { const P *ego = (const P *) ego_; INT i, n = ego->n, nb = ego->nb, is = ego->is, os = ego->os; R *w = ego->w, *W = ego->W; R *b = (R *) MALLOC(2 * nb * sizeof(R), BUFFERS); /* multiply input by conjugate bluestein sequence */ for (i = 0; i < n; ++i) { E xr = ri[i*is], xi = ii[i*is]; E wr = w[2*i], wi = w[2*i+1]; b[2*i] = xr * wr + xi * wi; b[2*i+1] = xi * wr - xr * wi; } for (; i < nb; ++i) b[2*i] = b[2*i+1] = K(0.0); /* convolution: FFT */ { plan_dft *cldf = (plan_dft *)ego->cldf; cldf->apply(ego->cldf, b, b+1, b, b+1); } /* convolution: pointwise multiplication */ for (i = 0; i < nb; ++i) { E xr = b[2*i], xi = b[2*i+1]; E wr = W[2*i], wi = W[2*i+1]; b[2*i] = xi * wr + xr * wi; b[2*i+1] = xr * wr - xi * wi; } /* convolution: IFFT by FFT with real/imag input/output swapped */ { plan_dft *cldf = (plan_dft *)ego->cldf; cldf->apply(ego->cldf, b, b+1, b, b+1); } /* multiply output by conjugate bluestein sequence */ for (i = 0; i < n; ++i) { E xi = b[2*i], xr = b[2*i+1]; E wr = w[2*i], wi = w[2*i+1]; ro[i*os] = xr * wr + xi * wi; io[i*os] = xi * wr - xr * wi; } X(ifree)(b); } static void awake(plan *ego_, enum wakefulness wakefulness) { P *ego = (P *) ego_; X(plan_awake)(ego->cldf, wakefulness); switch (wakefulness) { case SLEEPY: X(ifree0)(ego->w); ego->w = 0; X(ifree0)(ego->W); ego->W = 0; break; default: A(!ego->w); mktwiddle(wakefulness, ego); break; } } static int applicable(const solver *ego, const problem *p_, const planner *plnr) { const problem_dft *p = (const problem_dft *) p_; UNUSED(ego); return (1 && p->sz->rnk == 1 && p->vecsz->rnk == 0 /* FIXME: allow other sizes */ && X(is_prime)(p->sz->dims[0].n) /* FIXME: avoid infinite recursion of bluestein with itself. This works because all factors in child problems are 2, 3, 5 */ && p->sz->dims[0].n > 16 && CIMPLIES(NO_SLOWP(plnr), p->sz->dims[0].n > BLUESTEIN_MAX_SLOW) ); } static void destroy(plan *ego_) { P *ego = (P *) ego_; X(plan_destroy_internal)(ego->cldf); } static void print(const plan *ego_, printer *p) { const P *ego = (const P *)ego_; p->print(p, "(dft-bluestein-%D/%D%(%p%))", ego->n, ego->nb, ego->cldf); } static INT choose_transform_size(INT minsz) { while (!X(factors_into_small_primes)(minsz)) ++minsz; return minsz; } static plan *mkplan(const solver *ego, const problem *p_, planner *plnr) { const problem_dft *p = (const problem_dft *) p_; P *pln; INT n, nb; plan *cldf = 0; R *buf = (R *) 0; static const plan_adt padt = { X(dft_solve), awake, print, destroy }; if (!applicable(ego, p_, plnr)) return (plan *) 0; n = p->sz->dims[0].n; nb = choose_transform_size(2 * n - 1); buf = (R *) MALLOC(2 * nb * sizeof(R), BUFFERS); cldf = X(mkplan_f_d)(plnr, X(mkproblem_dft_d)(X(mktensor_1d)(nb, 2, 2), X(mktensor_1d)(1, 0, 0), buf, buf+1, buf, buf+1), NO_SLOW, 0, 0); if (!cldf) goto nada; X(ifree)(buf); pln = MKPLAN_DFT(P, &padt, apply); pln->n = n; pln->nb = nb; pln->w = 0; pln->W = 0; pln->cldf = cldf; pln->is = p->sz->dims[0].is; pln->os = p->sz->dims[0].os; X(ops_add)(&cldf->ops, &cldf->ops, &pln->super.super.ops); pln->super.super.ops.add += 4 * n + 2 * nb; pln->super.super.ops.mul += 8 * n + 4 * nb; pln->super.super.ops.other += 6 * (n + nb); return &(pln->super.super); nada: X(ifree0)(buf); X(plan_destroy_internal)(cldf); return (plan *)0; } static solver *mksolver(void) { static const solver_adt sadt = { PROBLEM_DFT, mkplan, 0 }; S *slv = MKSOLVER(S, &sadt); return &(slv->super); } void X(dft_bluestein_register)(planner *p) { REGISTER_SOLVER(p, mksolver()); }