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: #include "hc2hc.h"
Chris@10: 
Chris@10: hc2hc_solver *(*X(mksolver_hc2hc_hook))(size_t, INT, hc2hc_mkinferior) = 0;
Chris@10: 
Chris@10: typedef struct {
Chris@10:      plan_rdft super;
Chris@10:      plan *cld;
Chris@10:      plan *cldw;
Chris@10:      INT r;
Chris@10: } P;
Chris@10: 
Chris@10: static void apply_dit(const plan *ego_, R *I, R *O)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      plan_rdft *cld;
Chris@10:      plan_hc2hc *cldw;
Chris@10: 
Chris@10:      cld = (plan_rdft *) ego->cld;
Chris@10:      cld->apply(ego->cld, I, O);
Chris@10: 
Chris@10:      cldw = (plan_hc2hc *) ego->cldw;
Chris@10:      cldw->apply(ego->cldw, O);
Chris@10: }
Chris@10: 
Chris@10: static void apply_dif(const plan *ego_, R *I, R *O)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      plan_rdft *cld;
Chris@10:      plan_hc2hc *cldw;
Chris@10: 
Chris@10:      cldw = (plan_hc2hc *) ego->cldw;
Chris@10:      cldw->apply(ego->cldw, I);
Chris@10: 
Chris@10:      cld = (plan_rdft *) ego->cld;
Chris@10:      cld->apply(ego->cld, I, O);
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:      X(plan_awake)(ego->cldw, 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->cldw);
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, "(rdft-ct-%s/%D%(%p%)%(%p%))",
Chris@10: 	      ego->super.apply == apply_dit ? "dit" : "dif",
Chris@10: 	      ego->r, ego->cldw, ego->cld);
Chris@10: }
Chris@10: 
Chris@10: static int applicable0(const hc2hc_solver *ego, const problem *p_, planner *plnr)
Chris@10: {
Chris@10:      const problem_rdft *p = (const problem_rdft *) p_;
Chris@10:      INT r;
Chris@10: 
Chris@10:      return (1
Chris@10: 	     && p->sz->rnk == 1
Chris@10: 	     && p->vecsz->rnk <= 1 
Chris@10: 
Chris@10: 	     && (/* either the problem is R2HC, which is solved by DIT */
Chris@10: 		  (p->kind[0] == R2HC)
Chris@10: 		  ||
Chris@10: 		  /* or the problem is HC2R, in which case it is solved
Chris@10: 		     by DIF, which destroys the input */
Chris@10: 		  (p->kind[0] == HC2R && 
Chris@10: 		   (p->I == p->O || !NO_DESTROY_INPUTP(plnr))))
Chris@10: 		  
Chris@10: 	     && ((r = X(choose_radix)(ego->r, p->sz->dims[0].n)) > 0)
Chris@10: 	     && p->sz->dims[0].n > r);
Chris@10: }
Chris@10: 
Chris@10: int X(hc2hc_applicable)(const hc2hc_solver *ego, const problem *p_, planner *plnr)
Chris@10: {
Chris@10:      const problem_rdft *p;
Chris@10: 
Chris@10:      if (!applicable0(ego, p_, plnr))
Chris@10:           return 0;
Chris@10: 
Chris@10:      p = (const problem_rdft *) p_;
Chris@10: 
Chris@10:      return (0
Chris@10: 	     || p->vecsz->rnk == 0
Chris@10: 	     || !NO_VRECURSEP(plnr)
Chris@10: 	  );
Chris@10: }
Chris@10: 
Chris@10: static plan *mkplan(const solver *ego_, const problem *p_, planner *plnr)
Chris@10: {
Chris@10:      const hc2hc_solver *ego = (const hc2hc_solver *) ego_;
Chris@10:      const problem_rdft *p;
Chris@10:      P *pln = 0;
Chris@10:      plan *cld = 0, *cldw = 0;
Chris@10:      INT n, r, m, v, ivs, ovs;
Chris@10:      iodim *d;
Chris@10: 
Chris@10:      static const plan_adt padt = {
Chris@10: 	  X(rdft_solve), awake, print, destroy
Chris@10:      };
Chris@10: 
Chris@10:      if (NO_NONTHREADEDP(plnr) || !X(hc2hc_applicable)(ego, p_, plnr))
Chris@10:           return (plan *) 0;
Chris@10: 
Chris@10:      p = (const problem_rdft *) p_;
Chris@10:      d = p->sz->dims;
Chris@10:      n = d[0].n;
Chris@10:      r = X(choose_radix)(ego->r, n);
Chris@10:      m = n / r;
Chris@10: 
Chris@10:      X(tensor_tornk1)(p->vecsz, &v, &ivs, &ovs);
Chris@10: 
Chris@10:      switch (p->kind[0]) {
Chris@10: 	 case R2HC:
Chris@10: 	      cldw = ego->mkcldw(ego, 
Chris@10: 				 R2HC, r, m, d[0].os, v, ovs, 0, (m+2)/2, 
Chris@10: 				 p->O, plnr);
Chris@10: 	      if (!cldw) goto nada;
Chris@10: 
Chris@10: 	      cld = X(mkplan_d)(plnr, 
Chris@10: 				X(mkproblem_rdft_d)(
Chris@10: 				     X(mktensor_1d)(m, r * d[0].is, d[0].os),
Chris@10: 				     X(mktensor_2d)(r, d[0].is, m * d[0].os,
Chris@10: 						    v, ivs, ovs),
Chris@10: 				     p->I, p->O, p->kind)
Chris@10: 		   );
Chris@10: 	      if (!cld) goto nada;
Chris@10: 
Chris@10: 	      pln = MKPLAN_RDFT(P, &padt, apply_dit);
Chris@10: 	      break;
Chris@10: 
Chris@10: 	 case HC2R:
Chris@10: 	      cldw = ego->mkcldw(ego,
Chris@10: 				 HC2R, r, m, d[0].is, v, ivs, 0, (m+2)/2, 
Chris@10: 				 p->I, plnr);
Chris@10: 	      if (!cldw) goto nada;
Chris@10: 
Chris@10: 	      cld = X(mkplan_d)(plnr, 
Chris@10: 				X(mkproblem_rdft_d)(
Chris@10: 				     X(mktensor_1d)(m, d[0].is, r * d[0].os),
Chris@10: 				     X(mktensor_2d)(r, m * d[0].is, d[0].os,
Chris@10: 						    v, ivs, ovs),
Chris@10: 				     p->I, p->O, p->kind)
Chris@10: 		   );
Chris@10: 	      if (!cld) goto nada;
Chris@10: 	      
Chris@10: 	      pln = MKPLAN_RDFT(P, &padt, apply_dif);
Chris@10: 	      break;
Chris@10: 
Chris@10: 	 default: 
Chris@10: 	      A(0);
Chris@10:      }
Chris@10: 
Chris@10:      pln->cld = cld;
Chris@10:      pln->cldw = cldw;
Chris@10:      pln->r = r;
Chris@10:      X(ops_add)(&cld->ops, &cldw->ops, &pln->super.super.ops);
Chris@10: 
Chris@10:      /* inherit could_prune_now_p attribute from cldw */
Chris@10:      pln->super.super.could_prune_now_p = cldw->could_prune_now_p;
Chris@10: 
Chris@10:      return &(pln->super.super);
Chris@10: 
Chris@10:  nada:
Chris@10:      X(plan_destroy_internal)(cldw);
Chris@10:      X(plan_destroy_internal)(cld);
Chris@10:      return (plan *) 0;
Chris@10: }
Chris@10: 
Chris@10: hc2hc_solver *X(mksolver_hc2hc)(size_t size, INT r, hc2hc_mkinferior mkcldw)
Chris@10: {
Chris@10:      static const solver_adt sadt = { PROBLEM_RDFT, mkplan, 0 };
Chris@10:      hc2hc_solver *slv = (hc2hc_solver *)X(mksolver)(size, &sadt);
Chris@10:      slv->r = r;
Chris@10:      slv->mkcldw = mkcldw;
Chris@10:      return slv;
Chris@10: }
Chris@10: 
Chris@10: plan *X(mkplan_hc2hc)(size_t size, const plan_adt *adt, hc2hcapply apply)
Chris@10: {
Chris@10:      plan_hc2hc *ego;
Chris@10: 
Chris@10:      ego = (plan_hc2hc *) X(mkplan)(size, adt);
Chris@10:      ego->apply = apply;
Chris@10: 
Chris@10:      return &(ego->super);
Chris@10: }