Chris@10: /*
Chris@10:  * Copyright (c) 2003, 2007-11 Matteo Frigo
Chris@10:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
Chris@10:  *
Chris@10:  * This program is free software; you can redistribute it and/or modify
Chris@10:  * it under the terms of the GNU General Public License as published by
Chris@10:  * the Free Software Foundation; either version 2 of the License, or
Chris@10:  * (at your option) any later version.
Chris@10:  *
Chris@10:  * This program is distributed in the hope that it will be useful,
Chris@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@10:  * GNU General Public License for more details.
Chris@10:  *
Chris@10:  * You should have received a copy of the GNU General Public License
Chris@10:  * along with this program; if not, write to the Free Software
Chris@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@10:  *
Chris@10:  */
Chris@10: 
Chris@10: 
Chris@10: /* Solve an R2HC/HC2R problem via post/pre processing of a DHT.  This
Chris@10:    is mainly useful because we can use Rader to compute DHTs of prime
Chris@10:    sizes.  It also allows us to express hc2r problems in terms of r2hc
Chris@10:    (via dht-r2hc), and to do hc2r problems without destroying the input. */
Chris@10: 
Chris@10: #include "rdft.h"
Chris@10: 
Chris@10: typedef struct {
Chris@10:      solver super;
Chris@10: } S;
Chris@10: 
Chris@10: typedef struct {
Chris@10:      plan_rdft super;
Chris@10:      plan *cld;
Chris@10:      INT is, os;
Chris@10:      INT n;
Chris@10: } P;
Chris@10: 
Chris@10: static void apply_r2hc(const plan *ego_, R *I, R *O)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      INT os;
Chris@10:      INT i, n;
Chris@10: 
Chris@10:      {
Chris@10: 	  plan_rdft *cld = (plan_rdft *) ego->cld;
Chris@10: 	  cld->apply((plan *) cld, I, O);
Chris@10:      }
Chris@10: 
Chris@10:      n = ego->n;
Chris@10:      os = ego->os;
Chris@10:      for (i = 1; i < n - i; ++i) {
Chris@10: 	  E a, b;
Chris@10: 	  a = K(0.5) * O[os * i];
Chris@10: 	  b = K(0.5) * O[os * (n - i)];
Chris@10: 	  O[os * i] = a + b;
Chris@10: #if FFT_SIGN == -1
Chris@10: 	  O[os * (n - i)] = b - a;
Chris@10: #else
Chris@10: 	  O[os * (n - i)] = a - b;
Chris@10: #endif
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: /* hc2r, destroying input as usual */
Chris@10: static void apply_hc2r(const plan *ego_, R *I, R *O)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      INT is = ego->is;
Chris@10:      INT i, n = ego->n;
Chris@10: 
Chris@10:      for (i = 1; i < n - i; ++i) {
Chris@10: 	  E a, b;
Chris@10: 	  a = I[is * i];
Chris@10: 	  b = I[is * (n - i)];
Chris@10: #if FFT_SIGN == -1
Chris@10: 	  I[is * i] = a - b;
Chris@10: 	  I[is * (n - i)] = a + b;
Chris@10: #else
Chris@10: 	  I[is * i] = a + b;
Chris@10: 	  I[is * (n - i)] = a - b;
Chris@10: #endif
Chris@10:      }
Chris@10: 
Chris@10:      {
Chris@10: 	  plan_rdft *cld = (plan_rdft *) ego->cld;
Chris@10: 	  cld->apply((plan *) cld, I, O);
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: /* hc2r, without destroying input */
Chris@10: static void apply_hc2r_save(const plan *ego_, R *I, R *O)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      INT is = ego->is, os = ego->os;
Chris@10:      INT i, n = ego->n;
Chris@10: 
Chris@10:      O[0] = I[0];
Chris@10:      for (i = 1; i < n - i; ++i) {
Chris@10: 	  E a, b;
Chris@10: 	  a = I[is * i];
Chris@10: 	  b = I[is * (n - i)];
Chris@10: #if FFT_SIGN == -1
Chris@10: 	  O[os * i] = a - b;
Chris@10: 	  O[os * (n - i)] = a + b;
Chris@10: #else
Chris@10: 	  O[os * i] = a + b;
Chris@10: 	  O[os * (n - i)] = a - b;
Chris@10: #endif
Chris@10:      }
Chris@10:      if (i == n - i)
Chris@10: 	  O[os * i] = I[is * i];
Chris@10: 
Chris@10:      {
Chris@10: 	  plan_rdft *cld = (plan_rdft *) ego->cld;
Chris@10: 	  cld->apply((plan *) cld, O, O);
Chris@10:      }
Chris@10: }
Chris@10: 
Chris@10: static void awake(plan *ego_, enum wakefulness wakefulness)
Chris@10: {
Chris@10:      P *ego = (P *) ego_;
Chris@10:      X(plan_awake)(ego->cld, wakefulness);
Chris@10: }
Chris@10: 
Chris@10: static void destroy(plan *ego_)
Chris@10: {
Chris@10:      P *ego = (P *) ego_;
Chris@10:      X(plan_destroy_internal)(ego->cld);
Chris@10: }
Chris@10: 
Chris@10: static void print(const plan *ego_, printer *p)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      p->print(p, "(%s-dht-%D%(%p%))", 
Chris@10: 	      ego->super.apply == apply_r2hc ? "r2hc" : "hc2r",
Chris@10: 	      ego->n, ego->cld);
Chris@10: }
Chris@10: 
Chris@10: static int applicable0(const solver *ego_, const problem *p_)
Chris@10: {
Chris@10:      const problem_rdft *p = (const problem_rdft *) p_;
Chris@10:      UNUSED(ego_);
Chris@10: 
Chris@10:      return (1
Chris@10: 	     && p->sz->rnk == 1
Chris@10: 	     && p->vecsz->rnk == 0
Chris@10: 	     && (p->kind[0] == R2HC || p->kind[0] == HC2R)
Chris@10: 
Chris@10: 	     /* hack: size-2 DHT etc. are defined as being equivalent
Chris@10: 		to size-2 R2HC in problem.c, so we need this to prevent
Chris@10: 		infinite loops for size 2 in EXHAUSTIVE mode: */
Chris@10: 	     && p->sz->dims[0].n > 2
Chris@10: 	  );
Chris@10: }
Chris@10: 
Chris@10: static int applicable(const solver *ego, const problem *p_, 
Chris@10: 		      const planner *plnr)
Chris@10: {
Chris@10:      return (!NO_SLOWP(plnr) && applicable0(ego, p_));
Chris@10: }
Chris@10: 
Chris@10: static plan *mkplan(const solver *ego_, const problem *p_, planner *plnr)
Chris@10: {
Chris@10:      P *pln;
Chris@10:      const problem_rdft *p;
Chris@10:      problem *cldp;
Chris@10:      plan *cld;
Chris@10: 
Chris@10:      static const plan_adt padt = {
Chris@10: 	  X(rdft_solve), awake, print, destroy
Chris@10:      };
Chris@10: 
Chris@10:      if (!applicable(ego_, p_, plnr))
Chris@10:           return (plan *)0;
Chris@10: 
Chris@10:      p = (const problem_rdft *) p_;
Chris@10: 
Chris@10:      if (p->kind[0] == R2HC || !NO_DESTROY_INPUTP(plnr))
Chris@10: 	  cldp = X(mkproblem_rdft_1)(p->sz, p->vecsz, p->I, p->O, DHT);
Chris@10:      else {
Chris@10: 	  tensor *sz = X(tensor_copy_inplace)(p->sz, INPLACE_OS);
Chris@10: 	  cldp = X(mkproblem_rdft_1)(sz, p->vecsz, p->O, p->O, DHT);
Chris@10: 	  X(tensor_destroy)(sz);
Chris@10:      }
Chris@10:      cld = X(mkplan_d)(plnr, cldp);
Chris@10:      if (!cld) return (plan *)0;
Chris@10: 
Chris@10:      pln = MKPLAN_RDFT(P, &padt, p->kind[0] == R2HC ? 
Chris@10: 		       apply_r2hc : (NO_DESTROY_INPUTP(plnr) ?
Chris@10: 				     apply_hc2r_save : apply_hc2r));
Chris@10:      pln->n = p->sz->dims[0].n;
Chris@10:      pln->is = p->sz->dims[0].is;
Chris@10:      pln->os = p->sz->dims[0].os;
Chris@10:      pln->cld = cld;
Chris@10:      
Chris@10:      pln->super.super.ops = cld->ops;
Chris@10:      pln->super.super.ops.other += 4 * ((pln->n - 1)/2);
Chris@10:      pln->super.super.ops.add += 2 * ((pln->n - 1)/2);
Chris@10:      if (p->kind[0] == R2HC)
Chris@10: 	  pln->super.super.ops.mul += 2 * ((pln->n - 1)/2);
Chris@10:      if (pln->super.apply == apply_hc2r_save)
Chris@10: 	  pln->super.super.ops.other += 2 + (pln->n % 2 ? 0 : 2);
Chris@10: 
Chris@10:      return &(pln->super.super);
Chris@10: }
Chris@10: 
Chris@10: /* constructor */
Chris@10: static solver *mksolver(void)
Chris@10: {
Chris@10:      static const solver_adt sadt = { PROBLEM_RDFT, mkplan, 0 };
Chris@10:      S *slv = MKSOLVER(S, &sadt);
Chris@10:      return &(slv->super);
Chris@10: }
Chris@10: 
Chris@10: void X(rdft_dht_register)(planner *p)
Chris@10: {
Chris@10:      REGISTER_SOLVER(p, mksolver());
Chris@10: }