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: #include "mpi-rdft.h"
Chris@10: 
Chris@10: static void destroy(problem *ego_)
Chris@10: {
Chris@10:      problem_mpi_rdft *ego = (problem_mpi_rdft *) ego_;
Chris@10:      XM(dtensor_destroy)(ego->sz);
Chris@10:      MPI_Comm_free(&ego->comm);
Chris@10: #if !defined(STRUCT_HACK_C99) && !defined(STRUCT_HACK_KR)
Chris@10:      X(ifree0)(ego->kind);
Chris@10: #endif
Chris@10:      X(ifree)(ego_);
Chris@10: }
Chris@10: 
Chris@10: static void hash(const problem *p_, md5 *m)
Chris@10: {
Chris@10:      const problem_mpi_rdft *p = (const problem_mpi_rdft *) p_;
Chris@10:      int i;
Chris@10:      X(md5puts)(m, "mpi-dft");
Chris@10:      X(md5int)(m, p->I == p->O);
Chris@10:      /* don't include alignment -- may differ between processes
Chris@10: 	X(md5int)(m, X(alignment_of)(p->I));
Chris@10: 	X(md5int)(m, X(alignment_of)(p->O));
Chris@10: 	... note that applicability of MPI plans does not depend
Chris@10: 	    on alignment (although optimality may, in principle). */
Chris@10:      XM(dtensor_md5)(m, p->sz);
Chris@10:      X(md5INT)(m, p->vn);
Chris@10:      for (i = 0; i < p->sz->rnk; ++i)
Chris@10: 	  X(md5int)(m, p->kind[i]);
Chris@10:      X(md5int)(m, p->flags);
Chris@10:      MPI_Comm_size(p->comm, &i); X(md5int)(m, i);
Chris@10:      A(XM(md5_equal)(*m, p->comm));
Chris@10: }
Chris@10: 
Chris@10: static void print(const problem *ego_, printer *p)
Chris@10: {
Chris@10:      const problem_mpi_rdft *ego = (const problem_mpi_rdft *) ego_;
Chris@10:      int i;
Chris@10:      p->print(p, "(mpi-rdft %d %d %d ", 
Chris@10: 	      ego->I == ego->O,
Chris@10: 	      X(alignment_of)(ego->I),
Chris@10: 	      X(alignment_of)(ego->O));
Chris@10:      XM(dtensor_print)(ego->sz, p);
Chris@10:      for (i = 0; i < ego->sz->rnk; ++i)
Chris@10:           p->print(p, " %d", (int)ego->kind[i]);
Chris@10:      p->print(p, " %D %d", ego->vn, ego->flags);
Chris@10:      MPI_Comm_size(ego->comm, &i); p->print(p, " %d)", i);
Chris@10: }
Chris@10: 
Chris@10: static void zero(const problem *ego_)
Chris@10: {
Chris@10:      const problem_mpi_rdft *ego = (const problem_mpi_rdft *) ego_;
Chris@10:      R *I = ego->I;
Chris@10:      INT i, N;
Chris@10:      int my_pe;
Chris@10: 
Chris@10:      MPI_Comm_rank(ego->comm, &my_pe);
Chris@10:      N = ego->vn * XM(total_block)(ego->sz, IB, my_pe);
Chris@10:      for (i = 0; i < N; ++i) I[i] = K(0.0);
Chris@10: }
Chris@10: 
Chris@10: static const problem_adt padt =
Chris@10: {
Chris@10:      PROBLEM_MPI_RDFT,
Chris@10:      hash,
Chris@10:      zero,
Chris@10:      print,
Chris@10:      destroy
Chris@10: };
Chris@10: 
Chris@10: problem *XM(mkproblem_rdft)(const dtensor *sz, INT vn,
Chris@10: 			    R *I, R *O,
Chris@10: 			    MPI_Comm comm,
Chris@10: 			    const rdft_kind *kind, unsigned flags)
Chris@10: {
Chris@10:      problem_mpi_rdft *ego;
Chris@10:      int i, rnk = sz->rnk;
Chris@10:      int n_pes;
Chris@10: 
Chris@10:      A(XM(dtensor_validp)(sz) && FINITE_RNK(sz->rnk));
Chris@10:      MPI_Comm_size(comm, &n_pes);
Chris@10:      A(n_pes >= XM(num_blocks_total)(sz, IB)
Chris@10:        && n_pes >= XM(num_blocks_total)(sz, OB));
Chris@10:      A(vn >= 0);
Chris@10: 
Chris@10: #if defined(STRUCT_HACK_KR)
Chris@10:      ego = (problem_mpi_rdft *) X(mkproblem)(sizeof(problem_mpi_rdft)
Chris@10: 					     + sizeof(rdft_kind)
Chris@10: 					     * (rnk > 0 ? rnk - 1 : 0), &padt);
Chris@10: #elif defined(STRUCT_HACK_C99)
Chris@10:      ego = (problem_mpi_rdft *) X(mkproblem)(sizeof(problem_mpi_rdft)
Chris@10: 					     + sizeof(rdft_kind) * rnk, &padt);
Chris@10: #else
Chris@10:      ego = (problem_mpi_rdft *) X(mkproblem)(sizeof(problem_mpi_rdft), &padt);
Chris@10:      ego->kind = (rdft_kind *) MALLOC(sizeof(rdft_kind) * rnk, PROBLEMS);
Chris@10: #endif
Chris@10: 
Chris@10:      /* enforce pointer equality if untainted pointers are equal */
Chris@10:      if (UNTAINT(I) == UNTAINT(O))
Chris@10: 	  I = O = JOIN_TAINT(I, O);
Chris@10: 
Chris@10:      ego->sz = XM(dtensor_canonical)(sz, 0);
Chris@10:      ego->vn = vn;
Chris@10:      ego->I = I;
Chris@10:      ego->O = O;
Chris@10:      for (i = 0; i< ego->sz->rnk; ++i)
Chris@10: 	  ego->kind[i] = kind[i];
Chris@10: 
Chris@10:      /* canonicalize: replace TRANSPOSED_IN with TRANSPOSED_OUT by
Chris@10:         swapping the first two dimensions (for rnk > 1) */
Chris@10:      if ((flags & TRANSPOSED_IN) && ego->sz->rnk > 1) {
Chris@10: 	  rdft_kind k = ego->kind[0];
Chris@10: 	  ddim dim0 = ego->sz->dims[0];
Chris@10: 	  ego->sz->dims[0] = ego->sz->dims[1];
Chris@10: 	  ego->sz->dims[1] = dim0;
Chris@10: 	  ego->kind[0] = ego->kind[1];
Chris@10: 	  ego->kind[1] = k;
Chris@10: 	  flags &= ~TRANSPOSED_IN;
Chris@10: 	  flags ^= TRANSPOSED_OUT;
Chris@10:      }
Chris@10:      ego->flags = flags;
Chris@10: 
Chris@10:      MPI_Comm_dup(comm, &ego->comm);
Chris@10: 
Chris@10:      return &(ego->super);
Chris@10: }
Chris@10: 
Chris@10: problem *XM(mkproblem_rdft_d)(dtensor *sz, INT vn,
Chris@10: 			      R *I, R *O, 
Chris@10: 			      MPI_Comm comm,
Chris@10: 			      const rdft_kind *kind, unsigned flags)
Chris@10: {
Chris@10:      problem *p = XM(mkproblem_rdft)(sz, vn, I, O, comm, kind, flags);
Chris@10:      XM(dtensor_destroy)(sz);
Chris@10:      return p;
Chris@10: }