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 "fftw3-mpi.h"
Chris@42: #include "ifftw-mpi.h"
Chris@42: #include <string.h>
Chris@42: 
Chris@42: #if SIZEOF_SIZE_T == SIZEOF_UNSIGNED_INT
Chris@42: #  define FFTW_MPI_SIZE_T MPI_UNSIGNED
Chris@42: #elif SIZEOF_SIZE_T == SIZEOF_UNSIGNED_LONG
Chris@42: #  define FFTW_MPI_SIZE_T MPI_UNSIGNED_LONG
Chris@42: #elif SIZEOF_SIZE_T == SIZEOF_UNSIGNED_LONG_LONG
Chris@42: #  define FFTW_MPI_SIZE_T MPI_UNSIGNED_LONG_LONG
Chris@42: #else
Chris@42: #  error MPI type for size_t is unknown
Chris@42: #  define FFTW_MPI_SIZE_T MPI_UNSIGNED_LONG
Chris@42: #endif
Chris@42: 
Chris@42: /* Import wisdom from all processes to process 0, as prelude to
Chris@42:    exporting a single wisdom file (this is convenient when we are
Chris@42:    running on identical processors, to avoid the annoyance of having
Chris@42:    per-process wisdom files).  In order to make the time for this
Chris@42:    operation logarithmic in the number of processors (rather than
Chris@42:    linear), we employ a tree reduction algorithm.  This means that the
Chris@42:    wisdom is modified on processes other than root, which shouldn't
Chris@42:    matter in practice. */
Chris@42: void XM(gather_wisdom)(MPI_Comm comm_)
Chris@42: {
Chris@42:      MPI_Comm comm, comm2;
Chris@42:      int my_pe, n_pes;
Chris@42:      char *wis;
Chris@42:      size_t wislen;
Chris@42:      MPI_Status status;
Chris@42: 
Chris@42:      MPI_Comm_dup(comm_, &comm);
Chris@42:      MPI_Comm_rank(comm, &my_pe);
Chris@42:      MPI_Comm_size(comm, &n_pes);
Chris@42: 
Chris@42:      if (n_pes > 2) { /* recursively split into even/odd processes */
Chris@42: 	  MPI_Comm_split(comm, my_pe % 2, my_pe, &comm2);
Chris@42: 	  XM(gather_wisdom)(comm2);
Chris@42: 	  MPI_Comm_free(&comm2);
Chris@42:      }
Chris@42:      if (n_pes > 1 && my_pe < 2) { /* import process 1 -> 0 */
Chris@42: 	  if (my_pe == 1) {
Chris@42: 	       wis = X(export_wisdom_to_string)();
Chris@42: 	       wislen = strlen(wis) + 1;
Chris@42: 	       MPI_Send(&wislen, 1, FFTW_MPI_SIZE_T, 0, 111, comm);
Chris@42: 	       MPI_Send(wis, wislen, MPI_CHAR, 0, 222, comm);
Chris@42: 	       free(wis);
Chris@42: 	  }
Chris@42: 	  else /* my_pe == 0 */ {
Chris@42: 	       MPI_Recv(&wislen, 1, FFTW_MPI_SIZE_T, 1, 111, comm, &status);
Chris@42: 	       wis = (char *) MALLOC(wislen * sizeof(char), OTHER);
Chris@42: 	       MPI_Recv(wis, wislen, MPI_CHAR, 1, 222, comm, &status);
Chris@42: 	       if (!X(import_wisdom_from_string)(wis))
Chris@42: 		    MPI_Abort(comm, 1);
Chris@42: 	       X(ifree)(wis);
Chris@42: 	  }
Chris@42:      }
Chris@42:      MPI_Comm_free(&comm);
Chris@42: }
Chris@42: 
Chris@42: /* broadcast wisdom from process 0 to all other processes; this
Chris@42:    is useful so that we can import wisdom once and not worry
Chris@42:    about parallel I/O or process-specific wisdom, although of
Chris@42:    course it assumes that all the processes have identical
Chris@42:    performance characteristics (i.e. identical hardware). */
Chris@42: void XM(broadcast_wisdom)(MPI_Comm comm_)
Chris@42: {
Chris@42:      MPI_Comm comm;
Chris@42:      int my_pe;
Chris@42:      char *wis;
Chris@42:      size_t wislen;
Chris@42: 
Chris@42:      MPI_Comm_dup(comm_, &comm);
Chris@42:      MPI_Comm_rank(comm, &my_pe);
Chris@42: 
Chris@42:      if (my_pe != 0) {
Chris@42: 	  MPI_Bcast(&wislen, 1, FFTW_MPI_SIZE_T, 0, comm);
Chris@42: 	  wis = (char *) MALLOC(wislen * sizeof(char), OTHER);
Chris@42: 	  MPI_Bcast(wis, wislen, MPI_CHAR, 0, comm);
Chris@42: 	  if (!X(import_wisdom_from_string)(wis))
Chris@42: 	       MPI_Abort(comm, 1);
Chris@42: 	  X(ifree)(wis);
Chris@42:      }
Chris@42:      else /* my_pe == 0 */ {
Chris@42: 	  wis = X(export_wisdom_to_string)();
Chris@42: 	  wislen = strlen(wis) + 1;
Chris@42: 	  MPI_Bcast(&wislen, 1, FFTW_MPI_SIZE_T, 0, comm);
Chris@42: 	  MPI_Bcast(wis, wislen, MPI_CHAR, 0, comm);
Chris@42: 	  X(free)(wis);
Chris@42:      }
Chris@42:      MPI_Comm_free(&comm);
Chris@42: }