cannam@127: /*
cannam@127:  * Copyright (c) 2003, 2007-14 Matteo Frigo
cannam@127:  * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
cannam@127:  *
cannam@127:  * This program is free software; you can redistribute it and/or modify
cannam@127:  * it under the terms of the GNU General Public License as published by
cannam@127:  * the Free Software Foundation; either version 2 of the License, or
cannam@127:  * (at your option) any later version.
cannam@127:  *
cannam@127:  * This program is distributed in the hope that it will be useful,
cannam@127:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
cannam@127:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
cannam@127:  * GNU General Public License for more details.
cannam@127:  *
cannam@127:  * You should have received a copy of the GNU General Public License
cannam@127:  * along with this program; if not, write to the Free Software
cannam@127:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
cannam@127:  *
cannam@127:  */
cannam@127: 
cannam@127: #include "ct-hc2c.h"
cannam@127: #include "dft.h"
cannam@127: 
cannam@127: typedef struct {
cannam@127:      plan_rdft2 super;
cannam@127:      plan *cld;
cannam@127:      plan *cldw;
cannam@127:      INT r;
cannam@127: } P;
cannam@127: 
cannam@127: static void apply_dit(const plan *ego_, R *r0, R *r1, R *cr, R *ci)
cannam@127: {
cannam@127:      const P *ego = (const P *) ego_;
cannam@127:      plan_rdft *cld;
cannam@127:      plan_hc2c *cldw;
cannam@127:      UNUSED(r1);
cannam@127: 
cannam@127:      cld = (plan_rdft *) ego->cld;
cannam@127:      cld->apply(ego->cld, r0, cr);
cannam@127: 
cannam@127:      cldw = (plan_hc2c *) ego->cldw;
cannam@127:      cldw->apply(ego->cldw, cr, ci);
cannam@127: }
cannam@127: 
cannam@127: static void apply_dif(const plan *ego_, R *r0, R *r1, R *cr, R *ci)
cannam@127: {
cannam@127:      const P *ego = (const P *) ego_;
cannam@127:      plan_rdft *cld;
cannam@127:      plan_hc2c *cldw;
cannam@127:      UNUSED(r1);
cannam@127: 
cannam@127:      cldw = (plan_hc2c *) ego->cldw;
cannam@127:      cldw->apply(ego->cldw, cr, ci);
cannam@127: 
cannam@127:      cld = (plan_rdft *) ego->cld;
cannam@127:      cld->apply(ego->cld, cr, r0);
cannam@127: }
cannam@127: 
cannam@127: static void apply_dit_dft(const plan *ego_, R *r0, R *r1, R *cr, R *ci)
cannam@127: {
cannam@127:      const P *ego = (const P *) ego_;
cannam@127:      plan_dft *cld;
cannam@127:      plan_hc2c *cldw;
cannam@127: 
cannam@127:      cld = (plan_dft *) ego->cld;
cannam@127:      cld->apply(ego->cld, r0, r1, cr, ci);
cannam@127: 
cannam@127:      cldw = (plan_hc2c *) ego->cldw;
cannam@127:      cldw->apply(ego->cldw, cr, ci);
cannam@127: }
cannam@127: 
cannam@127: static void apply_dif_dft(const plan *ego_, R *r0, R *r1, R *cr, R *ci)
cannam@127: {
cannam@127:      const P *ego = (const P *) ego_;
cannam@127:      plan_dft *cld;
cannam@127:      plan_hc2c *cldw;
cannam@127: 
cannam@127:      cldw = (plan_hc2c *) ego->cldw;
cannam@127:      cldw->apply(ego->cldw, cr, ci);
cannam@127: 
cannam@127:      cld = (plan_dft *) ego->cld;
cannam@127:      cld->apply(ego->cld, ci, cr, r1, r0);
cannam@127: }
cannam@127: 
cannam@127: static void awake(plan *ego_, enum wakefulness wakefulness)
cannam@127: {
cannam@127:      P *ego = (P *) ego_;
cannam@127:      X(plan_awake)(ego->cld, wakefulness);
cannam@127:      X(plan_awake)(ego->cldw, wakefulness);
cannam@127: }
cannam@127: 
cannam@127: static void destroy(plan *ego_)
cannam@127: {
cannam@127:      P *ego = (P *) ego_;
cannam@127:      X(plan_destroy_internal)(ego->cldw);
cannam@127:      X(plan_destroy_internal)(ego->cld);
cannam@127: }
cannam@127: 
cannam@127: static void print(const plan *ego_, printer *p)
cannam@127: {
cannam@127:      const P *ego = (const P *) ego_;
cannam@127:      p->print(p, "(rdft2-ct-%s/%D%(%p%)%(%p%))",
cannam@127: 	      (ego->super.apply == apply_dit || 
cannam@127: 	       ego->super.apply == apply_dit_dft)
cannam@127: 	      ? "dit" : "dif",
cannam@127: 	      ego->r, ego->cldw, ego->cld);
cannam@127: }
cannam@127: 
cannam@127: static int applicable0(const hc2c_solver *ego, const problem *p_, planner *plnr)
cannam@127: {
cannam@127:      const problem_rdft2 *p = (const problem_rdft2 *) p_;
cannam@127:      INT r;
cannam@127: 
cannam@127:      return (1
cannam@127: 	     && p->sz->rnk == 1
cannam@127: 	     && p->vecsz->rnk <= 1 
cannam@127: 
cannam@127: 	     && (/* either the problem is R2HC, which is solved by DIT */
cannam@127: 		  (p->kind == R2HC)
cannam@127: 		  ||
cannam@127: 		  /* or the problem is HC2R, in which case it is solved
cannam@127: 		     by DIF, which destroys the input */
cannam@127: 		  (p->kind == HC2R && 
cannam@127: 		   (p->r0 == p->cr || !NO_DESTROY_INPUTP(plnr))))
cannam@127: 		  
cannam@127: 	     && ((r = X(choose_radix)(ego->r, p->sz->dims[0].n)) > 0)
cannam@127: 	     && p->sz->dims[0].n > r);
cannam@127: }
cannam@127: 
cannam@127: static int hc2c_applicable(const hc2c_solver *ego, const problem *p_,
cannam@127:                            planner *plnr)
cannam@127: {
cannam@127:      const problem_rdft2 *p;
cannam@127: 
cannam@127:      if (!applicable0(ego, p_, plnr))
cannam@127:           return 0;
cannam@127: 
cannam@127:      p = (const problem_rdft2 *) p_;
cannam@127: 
cannam@127:      return (0
cannam@127: 	     || p->vecsz->rnk == 0
cannam@127: 	     || !NO_VRECURSEP(plnr)
cannam@127: 	  );
cannam@127: }
cannam@127: 
cannam@127: static plan *mkplan(const solver *ego_, const problem *p_, planner *plnr)
cannam@127: {
cannam@127:      const hc2c_solver *ego = (const hc2c_solver *) ego_;
cannam@127:      const problem_rdft2 *p;
cannam@127:      P *pln = 0;
cannam@127:      plan *cld = 0, *cldw = 0;
cannam@127:      INT n, r, m, v, ivs, ovs;
cannam@127:      iodim *d;
cannam@127: 
cannam@127:      static const plan_adt padt = {
cannam@127: 	  X(rdft2_solve), awake, print, destroy
cannam@127:      };
cannam@127: 
cannam@127:      if (!hc2c_applicable(ego, p_, plnr))
cannam@127:           return (plan *) 0;
cannam@127: 
cannam@127:      p = (const problem_rdft2 *) p_;
cannam@127:      d = p->sz->dims;
cannam@127:      n = d[0].n;
cannam@127:      r = X(choose_radix)(ego->r, n);
cannam@127:      A((r % 2) == 0);
cannam@127:      m = n / r;
cannam@127: 
cannam@127:      X(tensor_tornk1)(p->vecsz, &v, &ivs, &ovs);
cannam@127: 
cannam@127:      switch (p->kind) {
cannam@127: 	 case R2HC:
cannam@127: 	      cldw = ego->mkcldw(ego, R2HC, 
cannam@127: 				 r, m * d[0].os, 
cannam@127: 				 m, d[0].os,
cannam@127: 				 v, ovs,
cannam@127: 				 p->cr, p->ci, plnr);
cannam@127: 	      if (!cldw) goto nada;
cannam@127: 
cannam@127: 	      switch (ego->hc2ckind) {
cannam@127: 		  case HC2C_VIA_RDFT:
cannam@127: 		       cld = X(mkplan_d)(
cannam@127: 			    plnr, 
cannam@127: 			    X(mkproblem_rdft_1_d)(
cannam@127: 				 X(mktensor_1d)(m, (r/2)*d[0].is, d[0].os),
cannam@127: 				 X(mktensor_3d)(
cannam@127: 				      2, p->r1 - p->r0, p->ci - p->cr,
cannam@127: 				      r / 2, d[0].is, m * d[0].os,
cannam@127: 				      v, ivs, ovs),
cannam@127: 				 p->r0, p->cr, R2HC) 
cannam@127: 			    );
cannam@127: 		       if (!cld) goto nada;
cannam@127: 
cannam@127: 		       pln = MKPLAN_RDFT2(P, &padt, apply_dit);
cannam@127: 		       break;
cannam@127: 
cannam@127: 		  case HC2C_VIA_DFT:
cannam@127: 		       cld = X(mkplan_d)(
cannam@127: 			    plnr, 
cannam@127: 			    X(mkproblem_dft_d)(
cannam@127: 				 X(mktensor_1d)(m, (r/2)*d[0].is, d[0].os),
cannam@127: 				 X(mktensor_2d)(
cannam@127: 				      r / 2, d[0].is, m * d[0].os,
cannam@127: 				      v, ivs, ovs),
cannam@127: 				 p->r0, p->r1, p->cr, p->ci) 
cannam@127: 			    );
cannam@127: 		       if (!cld) goto nada;
cannam@127: 
cannam@127: 		       pln = MKPLAN_RDFT2(P, &padt, apply_dit_dft);
cannam@127: 		       break;
cannam@127: 	      }
cannam@127: 	      break;
cannam@127: 
cannam@127: 	 case HC2R:
cannam@127: 	      cldw = ego->mkcldw(ego, HC2R, 
cannam@127: 				 r, m * d[0].is, 
cannam@127: 				 m, d[0].is,
cannam@127: 				 v, ivs,
cannam@127: 				 p->cr, p->ci, plnr);
cannam@127: 	      if (!cldw) goto nada;
cannam@127: 
cannam@127: 	      switch (ego->hc2ckind) {
cannam@127: 		  case HC2C_VIA_RDFT:
cannam@127: 		       cld = X(mkplan_d)(
cannam@127: 			    plnr, 
cannam@127: 			    X(mkproblem_rdft_1_d)(
cannam@127: 				 X(mktensor_1d)(m, d[0].is, (r/2)*d[0].os),
cannam@127: 				 X(mktensor_3d)(
cannam@127: 				      2, p->ci - p->cr, p->r1 - p->r0, 
cannam@127: 				      r / 2, m * d[0].is, d[0].os,
cannam@127: 				      v, ivs, ovs),
cannam@127: 				 p->cr, p->r0, HC2R) 
cannam@127: 			    );
cannam@127: 		       if (!cld) goto nada;
cannam@127: 
cannam@127: 		       pln = MKPLAN_RDFT2(P, &padt, apply_dif);
cannam@127: 		       break;
cannam@127: 
cannam@127: 		  case HC2C_VIA_DFT:
cannam@127: 		       cld = X(mkplan_d)(
cannam@127: 			    plnr, 
cannam@127: 			    X(mkproblem_dft_d)(
cannam@127: 				 X(mktensor_1d)(m, d[0].is, (r/2)*d[0].os),
cannam@127: 				 X(mktensor_2d)(
cannam@127: 				      r / 2, m * d[0].is, d[0].os,
cannam@127: 				      v, ivs, ovs),
cannam@127: 				 p->ci, p->cr, p->r1, p->r0) 
cannam@127: 			    );
cannam@127: 		       if (!cld) goto nada;
cannam@127: 
cannam@127: 		       pln = MKPLAN_RDFT2(P, &padt, apply_dif_dft);
cannam@127: 		       break;
cannam@127: 	      }
cannam@127: 	      break;
cannam@127: 
cannam@127: 	 default: 
cannam@127: 	      A(0);
cannam@127:      }
cannam@127: 
cannam@127:      pln->cld = cld;
cannam@127:      pln->cldw = cldw;
cannam@127:      pln->r = r;
cannam@127:      X(ops_add)(&cld->ops, &cldw->ops, &pln->super.super.ops);
cannam@127: 
cannam@127:      /* inherit could_prune_now_p attribute from cldw */
cannam@127:      pln->super.super.could_prune_now_p = cldw->could_prune_now_p;
cannam@127: 
cannam@127:      return &(pln->super.super);
cannam@127: 
cannam@127:  nada:
cannam@127:      X(plan_destroy_internal)(cldw);
cannam@127:      X(plan_destroy_internal)(cld);
cannam@127:      return (plan *) 0;
cannam@127: }
cannam@127: 
cannam@127: hc2c_solver *X(mksolver_hc2c)(size_t size, INT r, 
cannam@127: 			      hc2c_kind hc2ckind,
cannam@127: 			      hc2c_mkinferior mkcldw)
cannam@127: {
cannam@127:      static const solver_adt sadt = { PROBLEM_RDFT2, mkplan, 0 };
cannam@127:      hc2c_solver *slv = (hc2c_solver *)X(mksolver)(size, &sadt);
cannam@127:      slv->r = r;
cannam@127:      slv->hc2ckind = hc2ckind;
cannam@127:      slv->mkcldw = mkcldw;
cannam@127:      return slv;
cannam@127: }
cannam@127: 
cannam@127: plan *X(mkplan_hc2c)(size_t size, const plan_adt *adt, hc2capply apply)
cannam@127: {
cannam@127:      plan_hc2c *ego;
cannam@127: 
cannam@127:      ego = (plan_hc2c *) X(mkplan)(size, adt);
cannam@127:      ego->apply = apply;
cannam@127: 
cannam@127:      return &(ego->super);
cannam@127: }