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: /* Do a RODFT00 problem via an R2HC problem, padded antisymmetrically to
Chris@10:    twice the size.  This is asymptotically a factor of ~2 worse than
Chris@10:    rodft00e-r2hc.c (the algorithm used in e.g. FFTPACK and Numerical
Chris@10:    Recipes), but we abandoned the latter after we discovered that it
Chris@10:    has intrinsic accuracy problems. */
Chris@10: 
Chris@10: #include "reodft.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, *cldcpy;
Chris@10:      INT is;
Chris@10:      INT n;
Chris@10:      INT vl;
Chris@10:      INT ivs, ovs;
Chris@10: } P;
Chris@10: 
Chris@10: static void apply(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:      INT iv, vl = ego->vl;
Chris@10:      INT ivs = ego->ivs, ovs = ego->ovs;
Chris@10:      R *buf;
Chris@10: 
Chris@10:      buf = (R *) MALLOC(sizeof(R) * (2*n), BUFFERS);
Chris@10: 
Chris@10:      for (iv = 0; iv < vl; ++iv, I += ivs, O += ovs) {
Chris@10: 	  buf[0] = K(0.0);
Chris@10: 	  for (i = 1; i < n; ++i) {
Chris@10: 	       R a = I[(i-1) * is];
Chris@10: 	       buf[i] = -a;
Chris@10: 	       buf[2*n - i] = a;
Chris@10: 	  }
Chris@10: 	  buf[i] = K(0.0); /* i == n, Nyquist */
Chris@10: 	  
Chris@10: 	  /* r2hc transform of size 2*n */
Chris@10: 	  {
Chris@10: 	       plan_rdft *cld = (plan_rdft *) ego->cld;
Chris@10: 	       cld->apply((plan *) cld, buf, buf);
Chris@10: 	  }
Chris@10: 	  
Chris@10: 	  /* copy n-1 real numbers (imag. parts of hc array) from buf to O */
Chris@10: 	  {
Chris@10: 	       plan_rdft *cldcpy = (plan_rdft *) ego->cldcpy;
Chris@10: 	       cldcpy->apply((plan *) cldcpy, buf+2*n-1, O);
Chris@10: 	  }
Chris@10:      }
Chris@10: 
Chris@10:      X(ifree)(buf);
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->cldcpy, 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->cldcpy);
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, "(rodft00e-r2hc-pad-%D%v%(%p%)%(%p%))", 
Chris@10: 	      ego->n - 1, ego->vl, ego->cld, ego->cldcpy);
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:      return (1
Chris@10: 	     && p->sz->rnk == 1
Chris@10: 	     && p->vecsz->rnk <= 1
Chris@10: 	     && p->kind[0] == RODFT00
Chris@10: 	  );
Chris@10: }
Chris@10: 
Chris@10: static int applicable(const solver *ego, const problem *p, 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:      plan *cld = (plan *) 0, *cldcpy;
Chris@10:      R *buf = (R *) 0;
Chris@10:      INT n;
Chris@10:      INT vl, ivs, ovs;
Chris@10:      opcnt ops;
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: 	  goto nada;
Chris@10: 
Chris@10:      p = (const problem_rdft *) p_;
Chris@10: 
Chris@10:      n = p->sz->dims[0].n + 1;
Chris@10:      A(n > 0);
Chris@10:      buf = (R *) MALLOC(sizeof(R) * (2*n), BUFFERS);
Chris@10: 
Chris@10:      cld = X(mkplan_d)(plnr,X(mkproblem_rdft_1_d)(X(mktensor_1d)(2*n,1,1), 
Chris@10: 						  X(mktensor_0d)(), 
Chris@10: 						  buf, buf, R2HC));
Chris@10:      if (!cld)
Chris@10: 	  goto nada;
Chris@10: 
Chris@10:      X(tensor_tornk1)(p->vecsz, &vl, &ivs, &ovs);
Chris@10:      cldcpy =
Chris@10: 	  X(mkplan_d)(plnr,
Chris@10: 		      X(mkproblem_rdft_1_d)(X(mktensor_0d)(),
Chris@10: 					    X(mktensor_1d)(n-1,-1,
Chris@10: 							   p->sz->dims[0].os), 
Chris@10: 					    buf+2*n-1,TAINT(p->O, ovs), R2HC));
Chris@10:      if (!cldcpy)
Chris@10: 	  goto nada;
Chris@10: 
Chris@10:      X(ifree)(buf);
Chris@10: 
Chris@10:      pln = MKPLAN_RDFT(P, &padt, apply);
Chris@10: 
Chris@10:      pln->n = n;
Chris@10:      pln->is = p->sz->dims[0].is;
Chris@10:      pln->cld = cld;
Chris@10:      pln->cldcpy = cldcpy;
Chris@10:      pln->vl = vl;
Chris@10:      pln->ivs = ivs;
Chris@10:      pln->ovs = ovs;
Chris@10:      
Chris@10:      X(ops_zero)(&ops);
Chris@10:      ops.other = n-1 + 2*n; /* loads + stores (input -> buf) */
Chris@10: 
Chris@10:      X(ops_zero)(&pln->super.super.ops);
Chris@10:      X(ops_madd2)(pln->vl, &ops, &pln->super.super.ops);
Chris@10:      X(ops_madd2)(pln->vl, &cld->ops, &pln->super.super.ops);
Chris@10:      X(ops_madd2)(pln->vl, &cldcpy->ops, &pln->super.super.ops);
Chris@10: 
Chris@10:      return &(pln->super.super);
Chris@10: 
Chris@10:  nada:
Chris@10:      X(ifree0)(buf);
Chris@10:      if (cld)
Chris@10: 	  X(plan_destroy_internal)(cld);  
Chris@10:      return (plan *)0;
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(rodft00e_r2hc_pad_register)(planner *p)
Chris@10: {
Chris@10:      REGISTER_SOLVER(p, mksolver());
Chris@10: }