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: /* "MPI" DFTs where all of the data is on one processor...just
Chris@10:    call through to serial API. */
Chris@10: 
Chris@10: #include "mpi-rdft2.h"
Chris@10: #include "rdft.h"
Chris@10: 
Chris@10: typedef struct {
Chris@10:      plan_mpi_rdft2 super;
Chris@10:      plan *cld;
Chris@10:      INT vn;
Chris@10: } P;
Chris@10: 
Chris@10: static void apply_r2c(const plan *ego_, R *I, R *O)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      plan_rdft2 *cld;
Chris@10:      cld = (plan_rdft2 *) ego->cld;
Chris@10:      cld->apply(ego->cld, I, I+ego->vn, O, O+1);
Chris@10: }
Chris@10: 
Chris@10: static void apply_c2r(const plan *ego_, R *I, R *O)
Chris@10: {
Chris@10:      const P *ego = (const P *) ego_;
Chris@10:      plan_rdft2 *cld;
Chris@10:      cld = (plan_rdft2 *) ego->cld;
Chris@10:      cld->apply(ego->cld, O, O+ego->vn, I, I+1);
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: }
Chris@10: 
Chris@10: static void destroy(plan *ego_)
Chris@10: {
Chris@10:      P *ego = (P *) ego_;
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, "(mpi-rdft2-serial %(%p%))", ego->cld);
Chris@10: }
Chris@10: 
Chris@10: int XM(rdft2_serial_applicable)(const problem_mpi_rdft2 *p)
Chris@10: {
Chris@10:      return (1
Chris@10: 	     && p->flags == 0 /* TRANSPOSED/SCRAMBLED_IN/OUT not supported */
Chris@10: 	     && ((XM(is_local)(p->sz, IB) && XM(is_local)(p->sz, OB))
Chris@10: 		 || p->vn == 0));
Chris@10: }
Chris@10: 
Chris@10: static plan *mkplan(const solver *ego, const problem *p_, planner *plnr)
Chris@10: {
Chris@10:      const problem_mpi_rdft2 *p = (const problem_mpi_rdft2 *) p_;
Chris@10:      P *pln;
Chris@10:      plan *cld;
Chris@10:      int my_pe;
Chris@10:      R *r0, *r1, *cr, *ci;
Chris@10:      static const plan_adt padt = {
Chris@10:           XM(rdft2_solve), awake, print, destroy
Chris@10:      };
Chris@10: 
Chris@10:      UNUSED(ego);
Chris@10: 
Chris@10:      /* check whether applicable: */
Chris@10:      if (!XM(rdft2_serial_applicable)(p))
Chris@10:           return (plan *) 0;
Chris@10: 
Chris@10:      if (p->kind == R2HC) {
Chris@10: 	  r1 = (r0 = p->I) + p->vn;
Chris@10: 	  ci = (cr = p->O) + 1;
Chris@10:      }
Chris@10:      else {
Chris@10: 	  r1 = (r0 = p->O) + p->vn;
Chris@10: 	  ci = (cr = p->I) + 1;
Chris@10:      }
Chris@10: 
Chris@10:      MPI_Comm_rank(p->comm, &my_pe);
Chris@10:      if (my_pe == 0 && p->vn > 0) {
Chris@10: 	  INT ivs = 1 + (p->kind == HC2R), ovs = 1 + (p->kind == R2HC);
Chris@10: 	  int i, rnk = p->sz->rnk;
Chris@10: 	  tensor *sz = X(mktensor)(p->sz->rnk);
Chris@10: 	  sz->dims[rnk - 1].is = sz->dims[rnk - 1].os = 2 * p->vn;
Chris@10: 	  sz->dims[rnk - 1].n = p->sz->dims[rnk - 1].n / 2 + 1;
Chris@10: 	  for (i = rnk - 1; i > 0; --i) {
Chris@10: 	       sz->dims[i - 1].is = sz->dims[i - 1].os = 
Chris@10: 		    sz->dims[i].is * sz->dims[i].n;
Chris@10: 	       sz->dims[i - 1].n = p->sz->dims[i - 1].n;
Chris@10: 	  }
Chris@10: 	  sz->dims[rnk - 1].n = p->sz->dims[rnk - 1].n;
Chris@10: 
Chris@10: 	  cld = X(mkplan_d)(plnr,
Chris@10: 			    X(mkproblem_rdft2_d)(sz,
Chris@10: 						 X(mktensor_1d)(p->vn,ivs,ovs),
Chris@10: 						 r0, r1, cr, ci, p->kind));
Chris@10:      }
Chris@10:      else { /* idle process: make nop plan */
Chris@10: 	  cld = X(mkplan_d)(plnr,
Chris@10: 			    X(mkproblem_rdft2_d)(X(mktensor_0d)(),
Chris@10: 						 X(mktensor_1d)(0,0,0),
Chris@10: 						 cr, ci, cr, ci, HC2R));
Chris@10:      }
Chris@10:      if (XM(any_true)(!cld, p->comm)) return (plan *) 0;
Chris@10: 
Chris@10:      pln = MKPLAN_MPI_RDFT2(P, &padt, p->kind == R2HC ? apply_r2c : apply_c2r);
Chris@10:      pln->cld = cld;
Chris@10:      pln->vn = p->vn;
Chris@10:      X(ops_cpy)(&cld->ops, &pln->super.super.ops);
Chris@10:      return &(pln->super.super);
Chris@10: }
Chris@10: 
Chris@10: static solver *mksolver(void)
Chris@10: {
Chris@10:      static const solver_adt sadt = { PROBLEM_MPI_RDFT2, mkplan, 0 };
Chris@10:      return MKSOLVER(solver, &sadt);
Chris@10: }
Chris@10: 
Chris@10: void XM(rdft2_serial_register)(planner *p)
Chris@10: {
Chris@10:      REGISTER_SOLVER(p, mksolver());
Chris@10: }