Chris@42: /*
Chris@42:  * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@42:  * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@42:  *
Chris@42:  * This program is free software; you can redistribute it and/or modify
Chris@42:  * it under the terms of the GNU General Public License as published by
Chris@42:  * the Free Software Foundation; either version 2 of the License, or
Chris@42:  * (at your option) any later version.
Chris@42:  *
Chris@42:  * This program is distributed in the hope that it will be useful,
Chris@42:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@42:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@42:  * GNU General Public License for more details.
Chris@42:  *
Chris@42:  * You should have received a copy of the GNU General Public License
Chris@42:  * along with this program; if not, write to the Free Software
Chris@42:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@42:  *
Chris@42:  */
Chris@42: 
Chris@42: #include "threads.h"
Chris@42: 
Chris@42: typedef struct {
Chris@42:      plan_rdft super;
Chris@42:      plan *cld;
Chris@42:      plan **cldws;
Chris@42:      int nthr;
Chris@42:      INT r;
Chris@42: } P;
Chris@42: 
Chris@42: typedef struct {
Chris@42:      plan **cldws;
Chris@42:      R *IO;
Chris@42: } PD;
Chris@42: 
Chris@42: static void *spawn_apply(spawn_data *d)
Chris@42: {
Chris@42:      PD *ego = (PD *) d->data;
Chris@42:      
Chris@42:      plan_hc2hc *cldw = (plan_hc2hc *) (ego->cldws[d->thr_num]);
Chris@42:      cldw->apply((plan *) cldw, ego->IO);
Chris@42:      return 0;
Chris@42: }
Chris@42: 
Chris@42: static void apply_dit(const plan *ego_, R *I, R *O)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42:      plan_rdft *cld;
Chris@42: 
Chris@42:      cld = (plan_rdft *) ego->cld;
Chris@42:      cld->apply((plan *) cld, I, O);
Chris@42: 
Chris@42:      {
Chris@42: 	  PD d;
Chris@42: 	  
Chris@42: 	  d.IO = O;
Chris@42: 	  d.cldws = ego->cldws;
Chris@42: 
Chris@42: 	  X(spawn_loop)(ego->nthr, ego->nthr, spawn_apply, (void*)&d);
Chris@42:      }
Chris@42: }
Chris@42: 
Chris@42: static void apply_dif(const plan *ego_, R *I, R *O)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42:      plan_rdft *cld;
Chris@42: 
Chris@42:      {
Chris@42: 	  PD d;
Chris@42: 	  
Chris@42: 	  d.IO = I;
Chris@42: 	  d.cldws = ego->cldws;
Chris@42: 
Chris@42: 	  X(spawn_loop)(ego->nthr, ego->nthr, spawn_apply, (void*)&d);
Chris@42:      }
Chris@42: 
Chris@42:      cld = (plan_rdft *) ego->cld;
Chris@42:      cld->apply((plan *) cld, I, O);
Chris@42: }
Chris@42: 
Chris@42: static void awake(plan *ego_, enum wakefulness wakefulness)
Chris@42: {
Chris@42:      P *ego = (P *) ego_;
Chris@42:      int i;
Chris@42:      X(plan_awake)(ego->cld, wakefulness);
Chris@42:      for (i = 0; i < ego->nthr; ++i)
Chris@42: 	  X(plan_awake)(ego->cldws[i], wakefulness);
Chris@42: }
Chris@42: 
Chris@42: static void destroy(plan *ego_)
Chris@42: {
Chris@42:      P *ego = (P *) ego_;
Chris@42:      int i;
Chris@42:      X(plan_destroy_internal)(ego->cld);
Chris@42:      for (i = 0; i < ego->nthr; ++i)
Chris@42: 	  X(plan_destroy_internal)(ego->cldws[i]);
Chris@42:      X(ifree)(ego->cldws);
Chris@42: }
Chris@42: 
Chris@42: static void print(const plan *ego_, printer *p)
Chris@42: {
Chris@42:      const P *ego = (const P *) ego_;
Chris@42:      int i;
Chris@42:      p->print(p, "(rdft-thr-ct-%s-x%d/%D",
Chris@42: 	      ego->super.apply == apply_dit ? "dit" : "dif",
Chris@42: 	      ego->nthr, ego->r);
Chris@42:      for (i = 0; i < ego->nthr; ++i)
Chris@42:           if (i == 0 || (ego->cldws[i] != ego->cldws[i-1] &&
Chris@42:                          (i <= 1 || ego->cldws[i] != ego->cldws[i-2])))
Chris@42:                p->print(p, "%(%p%)", ego->cldws[i]);
Chris@42:      p->print(p, "%(%p%))", ego->cld);
Chris@42: }
Chris@42: 
Chris@42: static plan *mkplan(const solver *ego_, const problem *p_, planner *plnr)
Chris@42: {
Chris@42:      const hc2hc_solver *ego = (const hc2hc_solver *) ego_;
Chris@42:      const problem_rdft *p;
Chris@42:      P *pln = 0;
Chris@42:      plan *cld = 0, **cldws = 0;
Chris@42:      INT n, r, m, v, ivs, ovs, mcount;
Chris@42:      int i, nthr, plnr_nthr_save;
Chris@42:      INT block_size;
Chris@42:      iodim *d;
Chris@42: 
Chris@42:      static const plan_adt padt = {
Chris@42: 	  X(rdft_solve), awake, print, destroy
Chris@42:      };
Chris@42: 
Chris@42:      if (plnr->nthr <= 1 || !X(hc2hc_applicable)(ego, p_, plnr))
Chris@42:           return (plan *) 0;
Chris@42: 
Chris@42:      p = (const problem_rdft *) p_;
Chris@42:      d = p->sz->dims;
Chris@42:      n = d[0].n;
Chris@42:      r = X(choose_radix)(ego->r, n);
Chris@42:      m = n / r;
Chris@42:      mcount = (m + 2) / 2;
Chris@42: 
Chris@42:      X(tensor_tornk1)(p->vecsz, &v, &ivs, &ovs);
Chris@42: 
Chris@42:      block_size = (mcount + plnr->nthr - 1) / plnr->nthr;
Chris@42:      nthr = (int)((mcount + block_size - 1) / block_size);
Chris@42:      plnr_nthr_save = plnr->nthr;
Chris@42:      plnr->nthr = (plnr->nthr + nthr - 1) / nthr;
Chris@42: 
Chris@42:      cldws = (plan **) MALLOC(sizeof(plan *) * nthr, PLANS);
Chris@42:      for (i = 0; i < nthr; ++i) cldws[i] = (plan *) 0;
Chris@42: 
Chris@42:      switch (p->kind[0]) {
Chris@42: 	 case R2HC:
Chris@42: 	      for (i = 0; i < nthr; ++i) {
Chris@42: 		   cldws[i] = ego->mkcldw(ego, 
Chris@42: 					  R2HC, r, m, d[0].os, v, ovs, 
Chris@42: 					  i*block_size, 
Chris@42: 					  (i == nthr - 1) ? 
Chris@42: 					  (mcount - i*block_size) : block_size,
Chris@42: 					  p->O, plnr);
Chris@42: 		   if (!cldws[i]) goto nada;
Chris@42: 	      }
Chris@42: 
Chris@42: 	      plnr->nthr = plnr_nthr_save;
Chris@42: 
Chris@42: 	      cld = X(mkplan_d)(plnr, 
Chris@42: 				X(mkproblem_rdft_d)(
Chris@42: 				     X(mktensor_1d)(m, r * d[0].is, d[0].os),
Chris@42: 				     X(mktensor_2d)(r, d[0].is, m * d[0].os,
Chris@42: 						    v, ivs, ovs),
Chris@42: 				     p->I, p->O, p->kind)
Chris@42: 		   );
Chris@42: 	      if (!cld) goto nada;
Chris@42: 
Chris@42: 	      pln = MKPLAN_RDFT(P, &padt, apply_dit);
Chris@42: 	      break;
Chris@42: 
Chris@42: 	 case HC2R:
Chris@42: 	      for (i = 0; i < nthr; ++i) {
Chris@42: 		   cldws[i] = ego->mkcldw(ego, 
Chris@42: 					  HC2R, r, m, d[0].is, v, ivs, 
Chris@42: 					  i*block_size, 
Chris@42: 					  (i == nthr - 1) ? 
Chris@42: 					  (mcount - i*block_size) : block_size,
Chris@42: 					  p->I, plnr);
Chris@42: 		   if (!cldws[i]) goto nada;
Chris@42: 	      }
Chris@42: 
Chris@42: 	      plnr->nthr = plnr_nthr_save;
Chris@42: 
Chris@42: 	      cld = X(mkplan_d)(plnr, 
Chris@42: 				X(mkproblem_rdft_d)(
Chris@42: 				     X(mktensor_1d)(m, d[0].is, r * d[0].os),
Chris@42: 				     X(mktensor_2d)(r, m * d[0].is, d[0].os,
Chris@42: 						    v, ivs, ovs),
Chris@42: 				     p->I, p->O, p->kind)
Chris@42: 		   );
Chris@42: 	      if (!cld) goto nada;
Chris@42: 	      
Chris@42: 	      pln = MKPLAN_RDFT(P, &padt, apply_dif);
Chris@42: 	      break;
Chris@42: 
Chris@42: 	 default: 
Chris@42: 	      A(0);
Chris@42:      }
Chris@42: 
Chris@42:      pln->cld = cld;
Chris@42:      pln->cldws = cldws;
Chris@42:      pln->nthr = nthr;
Chris@42:      pln->r = r;
Chris@42:      X(ops_zero)(&pln->super.super.ops);
Chris@42:      for (i = 0; i < nthr; ++i) {
Chris@42:           X(ops_add2)(&cldws[i]->ops, &pln->super.super.ops);
Chris@42: 	  pln->super.super.could_prune_now_p |= cldws[i]->could_prune_now_p;
Chris@42:      }
Chris@42:      X(ops_add2)(&cld->ops, &pln->super.super.ops);
Chris@42:      return &(pln->super.super);
Chris@42: 
Chris@42:  nada:
Chris@42:      if (cldws) {
Chris@42: 	  for (i = 0; i < nthr; ++i)
Chris@42: 	       X(plan_destroy_internal)(cldws[i]);
Chris@42: 	  X(ifree)(cldws);
Chris@42:      }
Chris@42:      X(plan_destroy_internal)(cld);
Chris@42:      return (plan *) 0;
Chris@42: }
Chris@42: 
Chris@42: hc2hc_solver *X(mksolver_hc2hc_threads)(size_t size, INT r, 
Chris@42: 					hc2hc_mkinferior mkcldw)
Chris@42: {
Chris@42:      static const solver_adt sadt = { PROBLEM_RDFT, mkplan, 0 };
Chris@42:      hc2hc_solver *slv = (hc2hc_solver *)X(mksolver)(size, &sadt);
Chris@42:      slv->r = r;
Chris@42:      slv->mkcldw = mkcldw;
Chris@42:      return slv;
Chris@42: }