/*@
MPI_Group_incl - Produces a group by reordering an existing group and taking
only listed members
Input Parameters:
+ group - group (handle)
. n - number of elements in array 'ranks' (and size of newgroup) (integer)
- ranks - ranks of processes in 'group' to appear in 'newgroup' (array of
integers)
Output Parameters:
. newgroup - new group derived from above, in the order defined by 'ranks'
(handle)
.N ThreadSafe
.N Fortran
.N Errors
.N MPI_SUCCESS
.N MPI_ERR_GROUP
.N MPI_ERR_ARG
.N MPI_ERR_EXHAUSTED
.N MPI_ERR_RANK
.seealso: MPI_Group_free
@*/
int MPI_Group_incl(MPI_Group group, int n, const int ranks[], MPI_Group * newgroup)
{
int mpi_errno = MPI_SUCCESS;
MPIR_Group *group_ptr = NULL, *new_group_ptr = NULL;
MPIR_FUNC_TERSE_STATE_DECL(MPID_STATE_MPI_GROUP_INCL);
MPIR_ERRTEST_INITIALIZED_ORDIE();
MPID_THREAD_CS_ENTER(GLOBAL, MPIR_THREAD_GLOBAL_ALLFUNC_MUTEX);
MPIR_FUNC_TERSE_ENTER(MPID_STATE_MPI_GROUP_INCL);
/* Validate parameters, especially handles needing to be converted */
#ifdef HAVE_ERROR_CHECKING
{
MPID_BEGIN_ERROR_CHECKS;
{
MPIR_ERRTEST_GROUP(group, mpi_errno);
MPIR_ERRTEST_ARGNEG(n, "n", mpi_errno);
}
MPID_END_ERROR_CHECKS;
}
#endif
/* Convert MPI object handles to object pointers */
MPIR_Group_get_ptr(group, group_ptr);
/* Validate parameters and objects (post conversion) */
#ifdef HAVE_ERROR_CHECKING
{
MPID_BEGIN_ERROR_CHECKS;
{
/* Validate group_ptr */
MPIR_Group_valid_ptr(group_ptr, mpi_errno);
/* If group_ptr is not valid, it will be reset to null */
if (group_ptr) {
mpi_errno = MPIR_Group_check_valid_ranks(group_ptr, ranks, n);
}
if (mpi_errno)
goto fn_fail;
}
MPID_END_ERROR_CHECKS;
}
#endif /* HAVE_ERROR_CHECKING */
/* ... body of routine ... */
if (n == 0) {
*newgroup = MPI_GROUP_EMPTY;
goto fn_exit;
}
mpi_errno = MPIR_Group_incl_impl(group_ptr, n, ranks, &new_group_ptr);
if (mpi_errno)
goto fn_fail;
MPIR_OBJ_PUBLISH_HANDLE(*newgroup, new_group_ptr->handle);
/* ... end of body of routine ... */
fn_exit:
MPIR_FUNC_TERSE_EXIT(MPID_STATE_MPI_GROUP_INCL);
MPID_THREAD_CS_EXIT(GLOBAL, MPIR_THREAD_GLOBAL_ALLFUNC_MUTEX);
return mpi_errno;
fn_fail:
/* --BEGIN ERROR HANDLING-- */
#ifdef HAVE_ERROR_CHECKING
{
mpi_errno =
MPIR_Err_create_code(mpi_errno, MPIR_ERR_RECOVERABLE, __func__, __LINE__, MPI_ERR_OTHER,
"**mpi_group_incl", "**mpi_group_incl %G %d %p %p", group, n,
ranks, newgroup);
}
#endif
mpi_errno = MPIR_Err_return_comm(NULL, __func__, mpi_errno);
goto fn_exit;
/* --END ERROR HANDLING-- */
}
pami_result_t MPIDI_Comm_create_from_pami_geom(pami_geometry_range_t *task_slices,
size_t slice_count,
pami_geometry_t *geometry,
void **cookie)
{
int mpi_errno = MPI_SUCCESS;
int num_tasks = 0;
int *ranks = NULL;
MPID_Comm *comm_ptr = NULL, *new_comm_ptr = NULL;
MPID_Group *group_ptr = NULL, *new_group_ptr = NULL;
int i = 0, j = 0;
/* Get comm_ptr for MPI_COMM_WORLD and get the group_ptr for it */
MPID_Comm_get_ptr( MPI_COMM_WORLD, comm_ptr );
mpi_errno = MPIR_Comm_group_impl(comm_ptr, &group_ptr);
if (mpi_errno)
{
TRACE_ERR("Error while creating group_ptr from MPI_COMM_WORLD in MPIDI_Comm_create_from_pami_geom\n");
return PAMI_ERROR;
}
/* Create the ranks list from the pami_geometry_range_t array */
for(i = 0; i < slice_count; i++)
{
num_tasks += (task_slices[i].hi - task_slices[i].lo) + 1;
}
ranks = MPIU_Calloc0(num_tasks, int);
for(i = 0; i < slice_count; i++)
{
int slice_sz = (task_slices[i].hi - task_slices[i].lo) + 1;
int k = 0;
for(k = 0; k < slice_sz; k++)
{
ranks[j] = task_slices[i].lo + k;
j++;
}
}
/* Now we have all we need to create the new group. Create it */
mpi_errno = MPIR_Group_incl_impl(group_ptr, num_tasks, ranks, &new_group_ptr);
if (mpi_errno)
{
TRACE_ERR("Error while creating new_group_ptr from group_ptr in MPIDI_Comm_create_from_pami_geom\n");
return PAMI_ERROR;
}
/* Now create the communicator using the new_group_ptr */
mpi_errno = MPIR_Comm_create_intra(comm_ptr, new_group_ptr, &new_comm_ptr);
if (mpi_errno)
{
TRACE_ERR("Error while creating new_comm_ptr from group_ptr in MPIDI_Comm_create_from_pami_geom\n");
return PAMI_ERROR;
}
if(new_comm_ptr)
{
/* Get the geometry from the communicator and set the out parameters */
*geometry = new_comm_ptr->mpid.geometry;
*cookie = new_comm_ptr;
}
else
{
*geometry = PAMI_GEOMETRY_NULL;
*cookie = NULL;
}
/* Cleanup */
MPIU_TestFree(&ranks);
return PAMI_SUCCESS;
}
int MPIDI_CH3U_Get_failed_group(int last_rank, MPIR_Group **failed_group)
{
char *c;
int i, mpi_errno = MPI_SUCCESS, rank;
UT_array *failed_procs = NULL;
MPIR_Group *world_group;
MPIR_FUNC_VERBOSE_STATE_DECL(MPID_STATE_GET_FAILED_GROUP);
MPIR_FUNC_VERBOSE_ENTER(MPID_STATE_GET_FAILED_GROUP);
MPL_DBG_MSG_D(MPIDI_CH3_DBG_OTHER, VERBOSE, "Getting failed group with %d as last acknowledged\n", last_rank);
if (-1 == last_rank) {
MPL_DBG_MSG(MPIDI_CH3_DBG_OTHER, VERBOSE, "No failure acknowledged");
*failed_group = MPIR_Group_empty;
goto fn_exit;
}
if (*MPIDI_failed_procs_string == '\0') {
MPL_DBG_MSG(MPIDI_CH3_DBG_OTHER, VERBOSE, "Found no failed ranks");
*failed_group = MPIR_Group_empty;
goto fn_exit;
}
utarray_new(failed_procs, &ut_int_icd);
/* parse list of failed processes. This is a comma separated list
of ranks or ranges of ranks (e.g., "1, 3-5, 11") */
i = 0;
c = MPIDI_failed_procs_string;
while(1) {
parse_rank(&rank);
++i;
MPL_DBG_MSG_D(MPIDI_CH3_DBG_OTHER, VERBOSE, "Found failed rank: %d", rank);
utarray_push_back(failed_procs, &rank);
MPIDI_last_known_failed = rank;
MPIR_ERR_CHKINTERNAL(*c != ',' && *c != '\0', mpi_errno, "error parsing failed process list");
if (*c == '\0' || last_rank == rank)
break;
++c; /* skip ',' */
}
/* Create group of failed processes for comm_world. Failed groups for other
communicators can be created from this one using group_intersection. */
mpi_errno = MPIR_Comm_group_impl(MPIR_Process.comm_world, &world_group);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
mpi_errno = MPIR_Group_incl_impl(world_group, i, ut_int_array(failed_procs), failed_group);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
mpi_errno = MPIR_Group_release(world_group);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
fn_exit:
MPIR_FUNC_VERBOSE_EXIT(MPID_STATE_GET_FAILED_GROUP);
if (failed_procs)
utarray_free(failed_procs);
return mpi_errno;
fn_oom:
MPIR_ERR_SET1(mpi_errno, MPI_ERR_OTHER, "**nomem", "**nomem %s", "utarray");
fn_fail:
goto fn_exit;
}