Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: Re: [OMPI users] problem with groups and communicators in openmpi-1.6.4rc2
From: Ralph Castain (rhc_at_[hidden])
Date: 2013-01-19 09:44:08


I used your test code to confirm it also fails on our trunk - it looks like someone got the reference count wrong when creating/destructing groups.

Afraid I'll have to defer to the authors of that code area...

On Jan 19, 2013, at 1:27 AM, Siegmar Gross <Siegmar.Gross_at_[hidden]> wrote:

> Hi
>
> I have installed openmpi-1.6.4rc2 and have the following problem.
>
> tyr strided_vector 110 ompi_info | grep "Open MPI:"
> Open MPI: 1.6.4rc2r27861
> tyr strided_vector 111 mpicc -showme
> gcc -I/usr/local/openmpi-1.6.4_64_gcc/include -fexceptions -pthread -m64
> -L/usr/local/openmpi-1.6.4_64_gcc/lib64 -lmpi -lm -lkstat -llgrp -lsocket -lnsl
> -lrt -lm
>
>
> tyr strided_vector 112 mpiexec -np 4 data_type_4
> Process 2 of 4 running on tyr.informatik.hs-fulda.de
> Process 0 of 4 running on tyr.informatik.hs-fulda.de
> Process 3 of 4 running on tyr.informatik.hs-fulda.de
> Process 1 of 4 running on tyr.informatik.hs-fulda.de
>
> original matrix:
>
> 1 2 3 4 5 6 7 8 9 10
> 11 12 13 14 15 16 17 18 19 20
> 21 22 23 24 25 26 27 28 29 30
> 31 32 33 34 35 36 37 38 39 40
> 41 42 43 44 45 46 47 48 49 50
> 51 52 53 54 55 56 57 58 59 60
>
> result matrix:
> elements are sqared in columns:
> 0 1 2 6 7
> elements are multiplied with 2 in columns:
> 3 4 5 8 9
>
> 1 4 9 8 10 12 49 64 18 20
> 121 144 169 28 30 32 289 324 38 40
> 441 484 529 48 50 52 729 784 58 60
> 961 1024 1089 68 70 72 1369 1444 78 80
> 1681 1764 1849 88 90 92 2209 2304 98 100
> 2601 2704 2809 108 110 112 3249 3364 118 120
>
> Assertion failed: OPAL_OBJ_MAGIC_ID == ((opal_object_t *) (comm->c_remote_group)
> )->obj_magic_id, file ../../openmpi-1.6.4rc2r27861/ompi/communicator/comm_init.c
> , line 412
> [tyr:18578] *** Process received signal ***
> [tyr:18578] Signal: Abort (6)
> [tyr:18578] Signal code: (-1)
> Assertion failed: OPAL_OBJ_MAGIC_ID == ((opal_object_t *) (comm->c_remote_group)
> )->obj_magic_id, file ../../openmpi-1.6.4rc2r27861/ompi/communicator/comm_init.c
> , line 412
> /export2/prog/SunOS_sparc/openmpi-1.6.4_64_gcc/lib64/libmpi.so.1.0.7:opal_backtr
> ace_print+0x20
> [tyr:18580] *** Process received signal ***
> /export2/prog/SunOS_sparc/openmpi-1.6.4_64_gcc/lib64/libmpi.so.1.0.7:0x2c1bc4
> [tyr:18580] Signal: Abort (6)
> [tyr:18580] Signal code: (-1)
> /lib/sparcv9/libc.so.1:0xd88a4
> /lib/sparcv9/libc.so.1:0xcc418
> /lib/sparcv9/libc.so.1:0xcc624
> /lib/sparcv9/libc.so.1:__lwp_kill+0x8 [ Signal 6 (ABRT)]
> /lib/sparcv9/libc.so.1:abort+0xd0
> /lib/sparcv9/libc.so.1:_assert+0x74
> /export2/prog/SunOS_sparc/openmpi-1.6.4_64_gcc/lib64/libmpi.so.1.0.7:0xa4c58
> /export2/prog/SunOS_sparc/openmpi-1.6.4_64_gcc/lib64/libmpi.so.1.0.7:0xa2430
> /export2/prog/SunOS_sparc/openmpi-1.6.4_64_gcc/lib64/libmpi.so.1.0.7:ompi_comm_f
> inalize+0x168
> /export2/prog/SunOS_sparc/openmpi-1.6.4_64_gcc/lib64/libmpi.so.1.0.7:ompi_mpi_fi
> nalize+0xa60
> /export2/prog/SunOS_sparc/openmpi-1.6.4_64_gcc/lib64/libmpi.so.1.0.7:MPI_Finaliz
> e+0x90
> /home/fd1026/SunOS/sparc/bin/data_type_4:main+0x588
> /home/fd1026/SunOS/sparc/bin/data_type_4:_start+0x7c
> [tyr:18578] *** End of error message ***
> ...
>
>
>
> Everything works fine with LAM-MPI (even in a heterogeneous environment
> with little-endian and big-endian machines) so that it is probably an
> error in Open MPI (but you never know).
>
>
> tyr strided_vector 125 mpicc -showme
> gcc -I/usr/local/lam-6.5.9_64_gcc/include -L/usr/local/lam-6.5.9_64_gcc/lib
> -llamf77mpi -lmpi -llam -lsocket -lnsl
> tyr strided_vector 126 lamboot -v hosts.lam-mpi
>
> LAM 6.5.9/MPI 2 C++ - Indiana University
>
> Executing hboot on n0 (tyr.informatik.hs-fulda.de - 2 CPUs)...
> Executing hboot on n1 (sunpc1.informatik.hs-fulda.de - 4 CPUs)...
> topology done
>
> tyr strided_vector 127 mpirun -v app_data_type_4.lam-mpi
> 22894 data_type_4 running on local
> 22895 data_type_4 running on n0 (o)
> 21998 data_type_4 running on n1
> 22896 data_type_4 running on n0 (o)
> Process 1 of 4 running on tyr.informatik.hs-fulda.de
> Process 3 of 4 running on tyr.informatik.hs-fulda.de
> Process 2 of 4 running on sunpc1
> Process 0 of 4 running on tyr.informatik.hs-fulda.de
>
> original matrix:
>
> 1 2 3 4 5 6 7 8 9 10
> 11 12 13 14 15 16 17 18 19 20
> 21 22 23 24 25 26 27 28 29 30
> 31 32 33 34 35 36 37 38 39 40
> 41 42 43 44 45 46 47 48 49 50
> 51 52 53 54 55 56 57 58 59 60
>
> result matrix:
> elements are sqared in columns:
> 0 1 2 6 7
> elements are multiplied with 2 in columns:
> 3 4 5 8 9
>
> 1 4 9 8 10 12 49 64 18 20
> 121 144 169 28 30 32 289 324 38 40
> 441 484 529 48 50 52 729 784 58 60
> 961 1024 1089 68 70 72 1369 1444 78 80
> 1681 1764 1849 88 90 92 2209 2304 98 100
> 2601 2704 2809 108 110 112 3249 3364 118 120
>
> tyr strided_vector 128 lamhalt
>
> LAM 6.5.9/MPI 2 C++ - Indiana University
>
>
>
> I would be grateful, if somebody could fix the problem. Thank you
> very much for any help in advance.
>
>
> Kind regards
>
> Siegmar
> /* The program demonstrates how to set up and use a strided vector.
> * The process with rank 0 creates a matrix. The columns of the
> * matrix will then be distributed with a collective communication
> * operation to all processes. Each process performs an operation on
> * all column elements. Afterwards the results are collected in the
> * source matrix overwriting the original column elements.
> *
> * The program uses between one and n processes to change the values
> * of the column elements if the matrix has n columns. If you start
> * the program with one process it has to work on all n columns alone
> * and if you start it with n processes each process modifies the
> * values of one column. Every process must know how many columns it
> * has to modify so that it can allocate enough buffer space for its
> * column block. Therefore the process with rank 0 computes the
> * numbers of columns for each process in the array "num_columns" and
> * distributes this array with MPI_Broadcast to all processes. Each
> * process can now allocate memory for its column block. There is
> * still one task to do before the columns of the matrix can be
> * distributed with MPI_Scatterv: The size of every column block and
> * the offset of every column block must be computed und stored in
> * the arrays "sr_counts" and "sr_disps".
> *
> * An MPI data type is defined by its size, its contents, and its
> * extent. When multiple elements of the same size are used in a
> * contiguous manner (e.g. in a "scatter" operation or an operation
> * with "count" greater than one) the extent is used to compute where
> * the next element will start. The extent for a derived data type is
> * as big as the size of the derived data type so that the first
> * elements of the second structure will start after the last element
> * of the first structure, i.e., you have to "resize" the new data
> * type if you want to send it multiple times (count > 1) or to
> * scatter/gather it to many processes. Restrict the extent of the
> * derived data type for a strided vector in such a way that it looks
> * like just one element if it is used with "count > 1" or in a
> * scatter/gather operation.
> *
> * This version constructs a new column type (strided vector) with
> * "MPI_Type_vector" and uses collective communication. The new
> * data type knows the number of elements within one column and the
> * spacing between two column elements. The program uses at most
> * n processes if the matrix has n columns, i.e. depending on the
> * number of processes each process receives between 1 and n columns.
> * You can execute this program with an arbitrary number of processes
> * because it creates its own group with "num_worker" (<= n) processes
> * to perform the work if the matrix has n columns and the basic group
> * contains too many processes.
> *
> *
> * Compiling:
> * Store executable(s) into local directory.
> * mpicc -o <program name> <source code file name>
> *
> * Store executable(s) into predefined directories.
> * make
> *
> * Make program(s) automatically on all specified hosts. You must
> * edit the file "make_compile" and specify your host names before
> * you execute it.
> * make_compile
> *
> * Running:
> * LAM-MPI:
> * mpiexec -boot -np <number of processes> <program name>
> * or
> * mpiexec -boot \
> * -host <hostname> -np <number of processes> <program name> : \
> * -host <hostname> -np <number of processes> <program name>
> * or
> * mpiexec -boot [-v] -configfile <application file>
> * or
> * lamboot [-v] [<host file>]
> * mpiexec -np <number of processes> <program name>
> * or
> * mpiexec [-v] -configfile <application file>
> * lamhalt
> *
> * OpenMPI:
> * "host1", "host2", and so on can all have the same name,
> * if you want to start a virtual computer with some virtual
> * cpu's on the local host. The name "localhost" is allowed
> * as well.
> *
> * mpiexec -np <number of processes> <program name>
> * or
> * mpiexec --host <host1,host2,...> \
> * -np <number of processes> <program name>
> * or
> * mpiexec -hostfile <hostfile name> \
> * -np <number of processes> <program name>
> * or
> * mpiexec -app <application file>
> *
> * Cleaning:
> * local computer:
> * rm <program name>
> * or
> * make clean_all
> * on all specified computers (you must edit the file "make_clean_all"
> * and specify your host names before you execute it.
> * make_clean_all
> *
> *
> * File: data_type_4.c Author: S. Gross
> * Date: 30.08.2012
> *
> */
>
> #include <stdio.h>
> #include <stdlib.h>
> #include "mpi.h"
>
> #define P 6 /* # of rows */
> #define Q 10 /* # of columns */
> #define FACTOR 2 /* multiplicator for col. elem. */
> #define DEF_NUM_WORKER Q /* # of workers, must be <= Q */
>
> /* define macro to test the result of a "malloc" operation */
> #define TestEqualsNULL(val) \
> if (val == NULL) \
> { \
> fprintf (stderr, "file: %s line %d: Couldn't allocate memory.\n", \
> __FILE__, __LINE__); \
> exit (EXIT_FAILURE); \
> }
>
> /* define macro to determine the minimum of two values */
> #define MIN(a,b) ((a) < (b) ? (a) : (b))
>
>
> static void print_matrix (int p, int q, double **mat);
>
>
> int main (int argc, char *argv[])
> {
> int ntasks, /* number of parallel tasks */
> mytid, /* my task id */
> namelen, /* length of processor name */
> i, j, /* loop variables */
> *num_columns, /* # of columns in column block */
> *sr_counts, /* send/receive counts */
> *sr_disps, /* send/receive displacements */
> tmp, tmp1; /* temporary values */
> double matrix[P][Q],
> **col_block; /* column block of matrix */
> char processor_name[MPI_MAX_PROCESSOR_NAME];
> MPI_Datatype column_t, /* column type (strided vector) */
> col_block_t,
> tmp_column_t; /* needed to resize the extent */
> MPI_Group group_comm_world, /* processes in "basic group" */
> group_worker, /* processes in new groups */
> group_other;
> MPI_Comm COMM_WORKER, /* communicators for new groups */
> COMM_OTHER;
> int num_worker, /* # of worker in "group_worker"*/
> *group_w_mem, /* array of worker members */
> group_w_ntasks, /* # of tasks in "group_worker" */
> group_o_ntasks, /* # of tasks in "group_other" */
> group_w_mytid, /* my task id in "group_worker" */
> group_o_mytid, /* my task id in "group_other" */
> *universe_size_ptr, /* ptr to # of "virtual cpu's" */
> universe_size_flag; /* true if available */
>
> MPI_Init (&argc, &argv);
> MPI_Comm_rank (MPI_COMM_WORLD, &mytid);
> MPI_Comm_size (MPI_COMM_WORLD, &ntasks);
> /* Determine the correct number of processes for this program. If
> * there are more than Q processes (i.e., more processes than
> * columns) available, we split the "basic group" into two groups.
> * This program uses a group "group_worker" to do the real work
> * and a group "group_other" for the remaining processes of the
> * "basic group". The latter have nothing to do and can terminate
> * immediately. If there are less than or equal to Q processes
> * available all processes belong to group "group_worker" and group
> * "group_other" is empty. At first we find out which processes
> * belong to the "basic group".
> */
> MPI_Comm_group (MPI_COMM_WORLD, &group_comm_world);
> if (ntasks > Q)
> {
> /* There are too many processes, so that we must build a new group
> * with "num_worker" processes. "num_worker" will be the minimum of
> * DEF_NUM_WORKER and the "universe size" if it is supported by the
> * MPI implementation. At first we must check if DEF_NUM_WORKER has
> * a suitable value.
> */
> if (DEF_NUM_WORKER > Q)
> {
> if (mytid == 0)
> {
> fprintf (stderr, "\nError:\tInternal program error.\n"
> "\tConstant DEF_NUM_WORKER has value %d but must be\n"
> "\tlower than or equal to %d. Please change source\n"
> "\tcode and compile the program again.\n\n",
> DEF_NUM_WORKER, Q);
> }
> MPI_Group_free (&group_comm_world);
> MPI_Finalize ();
> exit (EXIT_FAILURE);
> }
> /* determine the universe size, set "num_worker" in an
> * appropriate way, and allocate memory for the array containing
> * the ranks of the members of the new group
> */
> MPI_Comm_get_attr (MPI_COMM_WORLD, MPI_UNIVERSE_SIZE,
> &universe_size_ptr, &universe_size_flag);
> if ((universe_size_flag != 0) && (*universe_size_ptr > 0))
> {
> num_worker = MIN (DEF_NUM_WORKER, *universe_size_ptr);
> }
> else
> {
> num_worker = DEF_NUM_WORKER;
> }
> group_w_mem = (int *) malloc (num_worker * sizeof (int));
> TestEqualsNULL (group_w_mem); /* test if memory was available */
> if (mytid == 0)
> {
> printf ("\nYou have started %d processes but I need at most "
> "%d processes.\n"
> "The universe contains %d \"virtual cpu's\" (\"0\" means "
> "not supported).\n"
> "I build a new worker group with %d processes. The "
> "processes with\n"
> "the following ranks in the basic group belong to "
> "the new group:\n ",
> ntasks, Q, *universe_size_ptr, num_worker);
> }
> for (i = 0; i < num_worker; ++i)
> {
> /* fetch some ranks from the basic group for the new worker
> * group, e.g. the last num_worker ranks to demonstrate that
> * a process may have different ranks in different groups
> */
> group_w_mem[i] = (ntasks - num_worker) + i;
> if (mytid == 0)
> {
> printf ("%d ", group_w_mem[i]);
> }
> }
> if (mytid == 0)
> {
> printf ("\n\n");
> }
> /* Create group "group_worker" */
> MPI_Group_incl (group_comm_world, num_worker, group_w_mem,
> &group_worker);
> free (group_w_mem);
> }
> else
> {
> /* there are at most as many processes as columns in our matrix,
> * i.e., we can use the "basic group"
> */
> group_worker = group_comm_world;
> }
> /* Create group "group_other" which demonstrates only how to use
> * another group operation and which has nothing to do in this
> * program.
> */
> MPI_Group_difference (group_comm_world, group_worker,
> &group_other);
> MPI_Group_free (&group_comm_world);
> /* Create communicators for both groups. The communicator is only
> * defined for all processes of the group and it is undefined
> * (MPI_COMM_NULL) for all other processes.
> */
> MPI_Comm_create (MPI_COMM_WORLD, group_worker, &COMM_WORKER);
> MPI_Comm_create (MPI_COMM_WORLD, group_other, &COMM_OTHER);
>
>
> /* =========================================================
> * ====== ======
> * ====== Supply work for all different groups. ======
> * ====== ======
> * ====== ======
> * ====== At first you must find out if a process ======
> * ====== belongs to a special group. You can use ======
> * ====== MPI_Group_rank for this purpose. It returns ======
> * ====== the rank of the calling process in the ======
> * ====== specified group or MPI_UNDEFINED if the ======
> * ====== calling process is not a member of the ======
> * ====== group. ======
> * ====== ======
> * =========================================================
> */
>
>
> /* =========================================================
> * ====== This is the group "group_worker". ======
> * =========================================================
> */
> MPI_Group_rank (group_worker, &group_w_mytid);
> if (group_w_mytid != MPI_UNDEFINED)
> {
> MPI_Comm_size (COMM_WORKER, &group_w_ntasks); /* # of processes */
> /* Now let's start with the real work */
> MPI_Get_processor_name (processor_name, &namelen);
> /* With the next statement every process executing this code will
> * print one line on the display. It may happen that the lines will
> * get mixed up because the display is a critical section. In general
> * only one process (mostly the process with rank 0) will print on
> * the display and all other processes will send their messages to
> * this process. Nevertheless for debugging purposes (or to
> * demonstrate that it is possible) it may be useful if every
> * process prints itself.
> */
> fprintf (stdout, "Process %d of %d running on %s\n",
> group_w_mytid, group_w_ntasks, processor_name);
> fflush (stdout);
> MPI_Barrier (COMM_WORKER); /* wait for all other processes */
>
> /* Build the new type for a strided vector and resize the extent
> * of the new datatype in such a way that the extent of the whole
> * column looks like just one element so that the next column
> * starts in matrix[0][i] in MPI_Scatterv/MPI_Gatherv.
> */
> MPI_Type_vector (P, 1, Q, MPI_DOUBLE, &tmp_column_t);
> MPI_Type_create_resized (tmp_column_t, 0, sizeof (double),
> &column_t);
> MPI_Type_commit (&column_t);
> MPI_Type_free (&tmp_column_t);
> if (group_w_mytid == 0)
> {
> tmp = 1;
> for (i = 0; i < P; ++i) /* initialize matrix */
> {
> for (j = 0; j < Q; ++j)
> {
> matrix[i][j] = tmp++;
> }
> }
> printf ("\n\noriginal matrix:\n\n");
> print_matrix (P, Q, (double **) matrix);
> }
> /* allocate memory for array containing the number of columns of a
> * column block for each process
> */
> num_columns = (int *) malloc (group_w_ntasks * sizeof (int));
> TestEqualsNULL (num_columns); /* test if memory was available */
>
> /* do an unnecessary initialization to make the GNU compiler happy
> * so that you won't get a warning about the use of a possibly
> * uninitialized variable
> */
> sr_counts = NULL;
> sr_disps = NULL;
> if (group_w_mytid == 0)
> {
> /* allocate memory for arrays containing the size and
> * displacement of each column block
> */
> sr_counts = (int *) malloc (group_w_ntasks * sizeof (int));
> TestEqualsNULL (sr_counts);
> sr_disps = (int *) malloc (group_w_ntasks * sizeof (int));
> TestEqualsNULL (sr_disps);
> /* compute number of columns in column block for each process */
> tmp = Q / group_w_ntasks;
> for (i = 0; i < group_w_ntasks; ++i)
> {
> num_columns[i] = tmp; /* number of columns */
> }
> for (i = 0; i < (Q % group_w_ntasks); ++i) /* adjust size */
> {
> num_columns[i]++;
> }
> for (i = 0; i < group_w_ntasks; ++i)
> {
> /* nothing to do because "column_t" contains already all
> * elements of a column, i.e., the "size" is equal to the
> * number of columns in the block
> */
> sr_counts[i] = num_columns[i]; /* "size" of column-block */
> }
> sr_disps[0] = 0; /* start of i-th column-block */
> for (i = 1; i < group_w_ntasks; ++i)
> {
> sr_disps[i] = sr_disps[i - 1] + sr_counts[i - 1];
> }
> }
> /* inform all processes about their column block sizes */
> MPI_Bcast (num_columns, group_w_ntasks, MPI_INT, 0, COMM_WORKER);
> /* allocate memory for a column block and define a new derived
> * data type for the column block. This data type is possibly
> * different for different processes if the number of processes
> * isn't a factor of the row size of the original matrix. Don't
> * forget to resize the extent of the new data type in such a
> * way that the extent of the whole column looks like just one
> * element so that the next column starts in col_block[0][i]
> * in MPI_Scatterv/MPI_Gatherv.
> */
> col_block = (double **) malloc (P * num_columns[group_w_mytid] *
> sizeof (double));
> TestEqualsNULL (col_block);
> MPI_Type_vector (P, 1, num_columns[group_w_mytid], MPI_DOUBLE,
> &tmp_column_t);
> MPI_Type_create_resized (tmp_column_t, 0, sizeof (double),
> &col_block_t);
> MPI_Type_commit (&col_block_t);
> MPI_Type_free (&tmp_column_t);
> /* send column block i of "matrix" to process i */
> MPI_Scatterv (matrix, sr_counts, sr_disps, column_t,
> col_block, num_columns[group_w_mytid],
> col_block_t, 0, COMM_WORKER);
> /* Modify column elements. The compiler doesn't know the structure
> * of the column block matrix so that you have to do the index
> * calculations for mat[i][j] yourself. In C a matrix is stored
> * row-by-row so that the i-th row starts at location "i * q" if
> * the matrix has "q" columns. Therefore the address of mat[i][j]
> * can be expressed as "(double *) mat + i * q + j" and mat[i][j]
> * itself as "*((double *) mat + i * q + j)".
> */
> for (i = 0; i < P; ++i)
> {
> for (j = 0; j < num_columns[group_w_mytid]; ++j)
> {
> if ((group_w_mytid % 2) == 0)
> {
> /* col_block[i][j] *= col_block[i][j] */
>
> *((double *) col_block + i * num_columns[group_w_mytid] + j) *=
> *((double *) col_block + i * num_columns[group_w_mytid] + j);
> }
> else
> {
> /* col_block[i][j] *= FACTOR */
>
> *((double *) col_block + i * num_columns[group_w_mytid] + j) *=
> FACTOR;
> }
> }
> }
> /* receive column-block i of "matrix" from process i */
> MPI_Gatherv (col_block, num_columns[group_w_mytid], col_block_t,
> matrix, sr_counts, sr_disps, column_t,
> 0, COMM_WORKER);
> if (group_w_mytid == 0)
> {
> printf ("\n\nresult matrix:\n"
> " elements are sqared in columns:\n ");
> tmp = 0;
> tmp1 = 0;
> for (i = 0; i < group_w_ntasks; ++i)
> {
> tmp1 = tmp1 + num_columns[i];
> if ((i % 2) == 0)
> {
> for (j = tmp; j < tmp1; ++j)
> {
> printf ("%4d", j);
> }
> }
> tmp = tmp1;
> }
> printf ("\n elements are multiplied with %d in columns:\n ",
> FACTOR);
> tmp = 0;
> tmp1 = 0;
> for (i = 0; i < group_w_ntasks; ++i)
> {
> tmp1 = tmp1 + num_columns[i];
> if ((i % 2) != 0)
> {
> for (j = tmp; j < tmp1; ++j)
> {
> printf ("%4d", j);
> }
> }
> tmp = tmp1;
> }
> printf ("\n\n\n");
> print_matrix (P, Q, (double **) matrix);
> free (sr_counts);
> free (sr_disps);
> }
> free (num_columns);
> free (col_block);
> MPI_Type_free (&column_t);
> MPI_Type_free (&col_block_t);
> MPI_Comm_free (&COMM_WORKER);
> }
>
>
> /* =========================================================
> * ====== This is the group "group_other". ======
> * =========================================================
> */
> MPI_Group_rank (group_other, &group_o_mytid);
> if (group_o_mytid != MPI_UNDEFINED)
> {
> /* Nothing to do (only to demonstrate how to divide work for
> * different groups).
> */
> MPI_Comm_size (COMM_OTHER, &group_o_ntasks);
> if (group_o_mytid == 0)
> {
> if (group_o_ntasks == 1)
> {
> printf ("\nGroup \"group_other\" contains %d process "
> "which has\n"
> "nothing to do.\n\n", group_o_ntasks);
> }
> else
> {
> printf ("\nGroup \"group_other\" contains %d processes "
> "which have\n"
> "nothing to do.\n\n", group_o_ntasks);
> }
> }
> MPI_Comm_free (&COMM_OTHER);
> }
>
>
> /* =========================================================
> * ====== all groups will reach this point ======
> * =========================================================
> */
> MPI_Group_free (&group_worker);
> MPI_Group_free (&group_other);
> MPI_Finalize ();
> return EXIT_SUCCESS;
> }
>
>
> /* Print the values of an arbitrary 2D-matrix of "double" values. The
> * compiler doesn't know the structure of the matrix so that you have
> * to do the index calculations for mat[i][j] yourself. In C a matrix
> * is stored row-by-row so that the i-th row starts at location "i * q"
> * if the matrix has "q" columns. Therefore the address of mat[i][j]
> * can be expressed as "(double *) mat + i * q + j" and mat[i][j]
> * itself as "*((double *) mat + i * q + j)".
> *
> * input parameters: p number of rows
> * q number of columns
> * mat 2D-matrix of "double" values
> * output parameters: none
> * return value: none
> * side effects: none
> *
> */
> void print_matrix (int p, int q, double **mat)
> {
> int i, j; /* loop variables */
>
> for (i = 0; i < p; ++i)
> {
> for (j = 0; j < q; ++j)
> {
> printf ("%6g", *((double *) mat + i * q + j));
> }
> printf ("\n");
> }
> printf ("\n");
> }
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users