Chris@82: /*
Chris@82:  * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@82:  * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@82:  *
Chris@82:  * This program is free software; you can redistribute it and/or modify
Chris@82:  * it under the terms of the GNU General Public License as published by
Chris@82:  * the Free Software Foundation; either version 2 of the License, or
Chris@82:  * (at your option) any later version.
Chris@82:  *
Chris@82:  * This program is distributed in the hope that it will be useful,
Chris@82:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@82:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@82:  * GNU General Public License for more details.
Chris@82:  *
Chris@82:  * You should have received a copy of the GNU General Public License
Chris@82:  * along with this program; if not, write to the Free Software
Chris@82:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@82:  *
Chris@82:  */
Chris@82: 
Chris@82: /* "MPI" RDFTs where all of the data is on one processor...just
Chris@82:    call through to serial API. */
Chris@82: 
Chris@82: #include "mpi-rdft.h"
Chris@82: 
Chris@82: typedef struct {
Chris@82:      plan_mpi_rdft super;
Chris@82:      plan *cld;
Chris@82: } P;
Chris@82: 
Chris@82: static void apply(const plan *ego_, R *I, R *O)
Chris@82: {
Chris@82:      const P *ego = (const P *) ego_;
Chris@82:      plan_rdft *cld = (plan_rdft *) ego->cld;
Chris@82:      cld->apply(ego->cld, I, O);
Chris@82: }
Chris@82: 
Chris@82: static void awake(plan *ego_, enum wakefulness wakefulness)
Chris@82: {
Chris@82:      P *ego = (P *) ego_;
Chris@82:      X(plan_awake)(ego->cld, wakefulness);
Chris@82: }
Chris@82: 
Chris@82: static void destroy(plan *ego_)
Chris@82: {
Chris@82:      P *ego = (P *) ego_;
Chris@82:      X(plan_destroy_internal)(ego->cld);
Chris@82: }
Chris@82: 
Chris@82: static void print(const plan *ego_, printer *p)
Chris@82: {
Chris@82:      const P *ego = (const P *) ego_;
Chris@82:      p->print(p, "(mpi-rdft-serial %(%p%))", ego->cld);
Chris@82: }
Chris@82: 
Chris@82: int XM(rdft_serial_applicable)(const problem_mpi_rdft *p)
Chris@82: {
Chris@82:      return (1
Chris@82: 	     && p->flags == 0 /* TRANSPOSED/SCRAMBLED_IN/OUT not supported */
Chris@82: 	     && ((XM(is_local)(p->sz, IB) && XM(is_local)(p->sz, OB))
Chris@82: 		 || p->vn == 0));
Chris@82: }
Chris@82: 
Chris@82: static plan *mkplan(const solver *ego, const problem *p_, planner *plnr)
Chris@82: {
Chris@82:      const problem_mpi_rdft *p = (const problem_mpi_rdft *) p_;
Chris@82:      P *pln;
Chris@82:      plan *cld;
Chris@82:      int my_pe;
Chris@82:      static const plan_adt padt = {
Chris@82:           XM(rdft_solve), awake, print, destroy
Chris@82:      };
Chris@82: 
Chris@82:      UNUSED(ego);
Chris@82: 
Chris@82:      /* check whether applicable: */
Chris@82:      if (!XM(rdft_serial_applicable)(p))
Chris@82:           return (plan *) 0;
Chris@82: 
Chris@82:      MPI_Comm_rank(p->comm, &my_pe);
Chris@82:      if (my_pe == 0 && p->vn > 0) {
Chris@82: 	  int i, rnk = p->sz->rnk;
Chris@82: 	  tensor *sz = X(mktensor)(rnk);
Chris@82: 	  rdft_kind *kind 
Chris@82: 	       = (rdft_kind *) MALLOC(sizeof(rdft_kind) * rnk, PROBLEMS);
Chris@82: 	  sz->dims[rnk - 1].is = sz->dims[rnk - 1].os = p->vn;
Chris@82: 	  sz->dims[rnk - 1].n = p->sz->dims[rnk - 1].n;
Chris@82: 	  for (i = rnk - 1; i > 0; --i) {
Chris@82: 	       sz->dims[i - 1].is = sz->dims[i - 1].os = 
Chris@82: 		    sz->dims[i].is * sz->dims[i].n;
Chris@82: 	       sz->dims[i - 1].n = p->sz->dims[i - 1].n;
Chris@82: 	  }
Chris@82: 	  for (i = 0; i < rnk; ++i)
Chris@82: 	       kind[i] = p->kind[i];
Chris@82: 	  
Chris@82: 	  cld = X(mkplan_d)(plnr,
Chris@82: 			    X(mkproblem_rdft_d)(sz,
Chris@82: 						X(mktensor_1d)(p->vn, 1, 1),
Chris@82: 						p->I, p->O, kind));
Chris@82: 	  X(ifree0)(kind);
Chris@82:      }
Chris@82:      else { /* idle process: make nop plan */
Chris@82: 	  cld = X(mkplan_d)(plnr,
Chris@82: 			    X(mkproblem_rdft_0_d)(X(mktensor_1d)(0,0,0),
Chris@82: 						  p->I, p->O));
Chris@82:      }
Chris@82:      if (XM(any_true)(!cld, p->comm)) return (plan *) 0;
Chris@82: 
Chris@82:      pln = MKPLAN_MPI_RDFT(P, &padt, apply);
Chris@82:      pln->cld = cld;
Chris@82:      X(ops_cpy)(&cld->ops, &pln->super.super.ops);
Chris@82:      return &(pln->super.super);
Chris@82: }
Chris@82: 
Chris@82: static solver *mksolver(void)
Chris@82: {
Chris@82:      static const solver_adt sadt = { PROBLEM_MPI_RDFT, mkplan, 0 };
Chris@82:      return MKSOLVER(solver, &sadt);
Chris@82: }
Chris@82: 
Chris@82: void XM(rdft_serial_register)(planner *p)
Chris@82: {
Chris@82:      REGISTER_SOLVER(p, mksolver());
Chris@82: }