1
1
openmpi/src/communicator/comm_dyn.c

672 строки
20 KiB
C
Исходник Обычный вид История

/*
* $HEADER$
*/
#include "ompi_config.h"
#include <string.h>
#include <stdio.h>
#ifdef HAVE_SYS_UIO_H
#include <sys/uio.h>
#endif
#include "mpi.h"
#include "communicator/communicator.h"
#include "datatype/datatype.h"
#include "errhandler/errhandler.h"
#include "proc/proc.h"
#include "threads/mutex.h"
#include "util/bit_ops.h"
#include "util/bufpack.h"
#include "util/argv.h"
#include "include/constants.h"
#include "mca/pcm/base/base.h"
#include "mca/pml/pml.h"
#include "mca/ns/base/base.h"
#include "mca/pml/pml.h"
#include "mca/oob/base/base.h"
#include "runtime/runtime.h"
#include "util/printf.h"
extern char **environ;
int ompi_comm_connect_accept ( ompi_communicator_t *comm, int root,
ompi_process_name_t *port, int send_first,
ompi_communicator_t **newcomm, int tag )
{
int size, rsize, rank, rc;
int namebuflen, rnamebuflen;
void *namebuf=NULL, *rnamebuf=NULL;
ompi_buffer_t sbuf;
ompi_buffer_t rbuf;
ompi_communicator_t *newcomp=MPI_COMM_NULL;
ompi_proc_t **rprocs=NULL;
ompi_group_t *group=comm->c_local_group;
ompi_process_name_t *rport=NULL;
ompi_buffer_t nbuf, nrbuf;
size = ompi_comm_size ( comm );
rank = ompi_comm_rank ( comm );
if ( rank == root ) {
/* The process receiving first does not have yet the contact
information of the remote process. Therefore, we have to
exchange that.
*/
if ( OMPI_COMM_JOIN_TAG != tag ) {
rport = ompi_comm_get_rport (port,send_first,
group->grp_proc_pointers[rank], tag);
}
else {
rport = port;
}
/* Exchange number of processes and msg length on both sides */
ompi_buffer_init (&nbuf, size*sizeof(ompi_process_name_t));
ompi_proc_get_namebuf (group->grp_proc_pointers, size, nbuf);
ompi_buffer_get(nbuf, &namebuf, &namebuflen);
ompi_buffer_init(&sbuf, 64);
ompi_pack(sbuf, &size, 1, OMPI_INT32);
ompi_pack(sbuf, &namebuflen, 1, OMPI_INT32);
if ( send_first ) {
rc = mca_oob_send_packed(rport, sbuf, tag, 0);
rc = mca_oob_recv_packed (rport, &rbuf, &tag);
}
else {
rc = mca_oob_recv_packed(rport, &rbuf, &tag);
rc = mca_oob_send_packed(rport, sbuf, tag, 0);
}
ompi_unpack(rbuf, &rsize, 1, OMPI_INT32);
ompi_unpack(rbuf, &rnamebuflen, 1, OMPI_INT32);
ompi_buffer_free(sbuf);
ompi_buffer_free(rbuf);
}
/* bcast the information to all processes in the local comm */
rc = comm->c_coll.coll_bcast (&rsize, 1, MPI_INT, root, comm );
if ( OMPI_SUCCESS != rc ) {
goto exit;
}
rc = comm->c_coll.coll_bcast (&rnamebuflen, 1, MPI_INT, root, comm );
if ( OMPI_SUCCESS != rc ) {
goto exit;
}
if ( rank == root ) {
/* Exchange list of processes in the groups */
if ( send_first ) {
rc = mca_oob_send_packed(rport, nbuf, tag, 0);
rc = mca_oob_recv_packed (rport, &nrbuf, &tag);
}
else {
rc = mca_oob_recv_packed(rport, &nrbuf, &tag);
rc = mca_oob_send_packed(rport, nbuf, tag, 0);
}
ompi_buffer_get(nrbuf, &rnamebuf, &rnamebuflen);
}
else {
/* non root processes need to allocate the buffer manually */
rnamebuf = (char *) malloc(rnamebuflen);
if ( NULL == rnamebuf ) {
rc = OMPI_ERR_OUT_OF_RESOURCE;
goto exit;
}
ompi_buffer_init_preallocated(&nrbuf, rnamebuf, rnamebuflen);
}
/* bcast list of processes to all procs in local group
and reconstruct the data. Note that proc_get_proclist
adds processes, which were not known yet to our
process pool.
*/
rc = comm->c_coll.coll_bcast (rnamebuf, rnamebuflen, MPI_BYTE, root, comm );
if ( OMPI_SUCCESS != rc ) {
goto exit;
}
rc = ompi_proc_get_proclist (nrbuf, rsize, &rprocs);
if ( OMPI_SUCCESS != rc ) {
goto exit;
}
ompi_buffer_free (nrbuf);
if ( rank == root ) {
ompi_buffer_free (nbuf);
}
/* allocate comm-structure */
newcomp = ompi_comm_allocate ( size, rsize );
if ( NULL == newcomp ) {
rc = OMPI_ERR_OUT_OF_RESOURCE;
goto exit;
}
/* allocate comm_cid */
rc = ompi_comm_nextcid ( newcomp, /* new communicator */
comm, /* old communicator */
NULL, /* bridge comm */
&root, /* local leader */
rport, /* remote leader */
OMPI_COMM_CID_INTRA_OOB, /* mode */
send_first ); /* send or recv first */
if ( OMPI_SUCCESS != rc ) {
goto exit;
}
/* set up communicator structure */
rc = ompi_comm_set ( newcomp, /* new comm */
comm, /* old comm */
group->grp_proc_count, /* local_size */
group->grp_proc_pointers, /* local_procs*/
rsize, /* remote_size */
rprocs, /* remote_procs */
NULL, /* attrs */
comm->error_handler, /* error handler */
NULL /* topo component */
);
/* activate comm and init coll-component */
rc = ompi_comm_activate ( newcomp, /* new communicator */
comm, /* old communicator */
NULL, /* bridge comm */
&root, /* local leader */
rport, /* remote leader */
OMPI_COMM_CID_INTRA_OOB, /* mode */
send_first, /* send or recv first */
NULL ); /* coll component */
if ( OMPI_SUCCESS != rc ) {
goto exit;
}
/* Question: do we have to re-start some low level stuff
to enable the usage of fast communication devices
between the two worlds ?
*/
exit:
if ( NULL != rprocs ) {
free ( rprocs );
}
if ( OMPI_SUCCESS != rc ) {
if ( MPI_COMM_NULL != newcomp ) {
OBJ_RETAIN(newcomp);
newcomp = MPI_COMM_NULL;
}
}
*newcomm = newcomp;
return rc;
}
/**********************************************************************/
/**********************************************************************/
/**********************************************************************/
/*
* This routine is necessary, since in the connect/accept case, the processes
* executing the connect operation have the OOB contact information of the
* leader of the remote group, however, the processes executing the
* accept get their own port_name = OOB contact information passed in as
* an argument. This is however useless.
*
* Therefore, the two root processes exchange this information at this point.
*
*/
ompi_process_name_t *ompi_comm_get_rport (ompi_process_name_t *port, int send_first,
ompi_proc_t *proc, int tag)
{
int rc;
ompi_process_name_t *rport, tbuf;
ompi_proc_t *rproc=NULL;
bool isnew = false;
if ( send_first ) {
ompi_buffer_t sbuf;
rproc = ompi_proc_find_and_add(port, &isnew);
ompi_buffer_init(&sbuf, sizeof(ompi_process_name_t));
ompi_pack(sbuf, &(proc->proc_name), 1, OMPI_NAME);
rc = mca_oob_send_packed(port, sbuf, tag, 0);
ompi_buffer_free(sbuf);
rport = port;
}
else {
ompi_buffer_t rbuf;
rc = mca_oob_recv_packed(MCA_OOB_NAME_ANY, &rbuf, &tag);
ompi_unpack(rbuf, &tbuf, 1, OMPI_NAME);
ompi_buffer_free(rbuf);
rproc = ompi_proc_find_and_add(&tbuf, &isnew);
rport = &(rproc->proc_name);
}
if(isnew)
mca_pml.pml_add_procs(&rproc, 1);
return rport;
}
/**********************************************************************/
/**********************************************************************/
/**********************************************************************/
int ompi_comm_start_processes (char *command, char **argv, int maxprocs,
MPI_Info info, char *port_name )
{
mca_ns_base_jobid_t new_jobid;
ompi_rte_node_schedule_t *sched;
ompi_rte_spawn_handle_t *spawn_handle;
ompi_list_t *nodelist=NULL;
ompi_list_t schedlist;
char *tmp, *envvarname, *segment, *my_contact_info;
char cwd[MAXPATHLEN];
int rc;
/* parse the info object */
/* check potentially for:
- "host": desired host where to spawn the processes
- "arch": desired architecture
- "wdir": directory, where executable can be found
- "path": list of directories where to look for the executable
- "file": filename, where additional information is provided.
- "soft": see page 92 of MPI-2.
*/
/* get the jobid for the new processes */
new_jobid = ompi_name_server.create_jobid();
/* get the spawn handle to start spawning stuff */
spawn_handle =
ompi_rte_get_spawn_handle(OMPI_RTE_SPAWN_FROM_MPI|OMPI_RTE_SPAWN_HIGH_QOS,
true);
if (NULL == spawn_handle) {
printf("show_help: get_spawn_handle failed\n");
return -1;
}
/* BWB - fix jobid, procs, and nodes */
nodelist = ompi_rte_allocate_resources(spawn_handle, new_jobid, 0, maxprocs);
if (NULL == nodelist) {
/* BWB show_help */
printf("show_help: ompi_rte_allocate_resources failed\n");
return -1;
}
/*
* Process mapping
*/
OBJ_CONSTRUCT(&schedlist, ompi_list_t);
sched = OBJ_NEW(ompi_rte_node_schedule_t);
ompi_list_append(&schedlist, (ompi_list_item_t*) sched);
/* ompi_cmd_line_get_tail(cmd_line, &(sched->argc), &(sched->argv)); */
ompi_argv_append (&(sched->argc), &(sched->argv), command);
if (argv != MPI_ARGV_NULL ) {
int i=0;
char *arg=argv[i];
while ( arg!=NULL ) {
ompi_argv_append(&(sched->argc), &(sched->argv), arg);
arg = argv[++i];
}
}
/*
* build environment to be passed
*/
mca_pcm_base_build_base_env(environ, &(sched->envc), &(sched->env));
/* set initial contact info */
my_contact_info = strdup(ompi_universe_info.ns_replica);
asprintf(&tmp, "OMPI_MCA_ns_base_replica=%s", my_contact_info);
ompi_argv_append(&(sched->envc), &(sched->env), tmp);
free(tmp);
asprintf(&tmp, "OMPI_MCA_gpr_base_replica=%s", my_contact_info);
ompi_argv_append(&(sched->envc), &(sched->env), tmp);
free(tmp);
if (NULL != ompi_universe_info.name) {
asprintf(&tmp, "OMPI_universe_name=%s", ompi_universe_info.name);
ompi_argv_append(&(sched->envc), &(sched->env), tmp);
free(tmp);
}
/* Add environment variable with the contact information for the
child processes */
asprintf(&envvarname, "OMPI_PARENT_PORT_%u", new_jobid);
asprintf(&tmp, "%s=%s", envvarname, port_name);
ompi_argv_append(&(sched->envc), &(sched->env), tmp);
free(tmp);
free(envvarname);
getcwd(cwd, MAXPATHLEN);
sched->cwd = strdup(cwd);
sched->nodelist = nodelist;
if (sched->argc == 0) {
printf("no app to start\n");
return MPI_ERR_ARG;
}
/*
* register to monitor the startup and shutdown processes
*/
/* setup segment for this job */
asprintf(&segment, "ompi-job-%d", new_jobid);
/* register a synchro on the segment so we get notified when everyone registers */
rc = ompi_registry.synchro(
OMPI_REGISTRY_SYNCHRO_MODE_LEVEL|OMPI_REGISTRY_SYNCHRO_MODE_ONE_SHOT,
OMPI_REGISTRY_OR,
segment,
NULL,
maxprocs,
ompi_rte_all_procs_registered, NULL);
/*
* spawn procs
*/
if (OMPI_SUCCESS != ompi_rte_spawn_procs(spawn_handle, new_jobid, &schedlist)) {
printf("show_help: woops! we didn't spawn :( \n");
return MPI_ERR_SPAWN;
}
if (OMPI_SUCCESS != ompi_rte_monitor_procs_registered()) {
printf("procs didn't all register - returning an error\n");
return MPI_ERR_SPAWN;
}
/*
* Clean up
*/
if (NULL != nodelist) ompi_rte_deallocate_resources(spawn_handle,
new_jobid, nodelist);
if (NULL != spawn_handle) OBJ_RELEASE(spawn_handle);
OBJ_DESTRUCT(&schedlist);
return OMPI_SUCCESS;
}
/**********************************************************************/
/**********************************************************************/
/**********************************************************************/
int ompi_comm_dyn_init (void)
{
uint32_t jobid;
size_t size;
ompi_proc_t **myproc=NULL;
char *envvarname=NULL, *port_name=NULL;
char *oob_port=NULL;
int tag, root=0, send_first=1;
ompi_communicator_t *newcomm=NULL;
ompi_process_name_t *port_proc_name=NULL;
/* get jobid */
myproc = ompi_proc_self(&size);
jobid = ompi_name_server.get_jobid(&(myproc[0]->proc_name));
/* check for appropriate env variable */
asprintf(&envvarname, "OMPI_PARENT_PORT_%u", jobid);
port_name = getenv(envvarname);
free (envvarname);
/* if env-variable is set, parse port and call comm_connect_accept */
if (NULL != port_name ) {
/* we have been spawned */
oob_port = ompi_parse_port (port_name, &tag);
port_proc_name = ompi_name_server.convert_string_to_process_name(oob_port);
ompi_comm_connect_accept (MPI_COMM_WORLD, root, port_proc_name,
send_first, &newcomm, tag );
/* Set the parent communicator */
ompi_mpi_comm_parent = newcomm;
/* originally, we set comm_parent to comm_null (in comm_init),
* now we have to decrease the reference counters to the according
* objects
*/
OBJ_RETAIN(&ompi_mpi_comm_null);
OBJ_RETAIN(&ompi_mpi_group_null);
OBJ_RETAIN(&ompi_mpi_group_null);
OBJ_RETAIN(&ompi_mpi_errors_are_fatal);
/* Set name for debugging purposes */
snprintf(newcomm->c_name, MPI_MAX_OBJECT_NAME, "MPI_COMM_PARENT");
}
return OMPI_SUCCESS;
}
/**********************************************************************/
/**********************************************************************/
/**********************************************************************/
/* this routine runs through the list of communicators and
and does the disconnect for all dynamic communicators */
int ompi_comm_dyn_finalize (void)
{
int i,j=0, max=0;
ompi_comm_disconnect_obj **objs=NULL;
ompi_communicator_t *comm=NULL;
if ( 1 <ompi_comm_num_dyncomm ) {
objs = (ompi_comm_disconnect_obj **)malloc (ompi_comm_num_dyncomm*
sizeof(ompi_comm_disconnect_obj*));
if ( NULL == objs ) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
max = ompi_pointer_array_get_size(&ompi_mpi_communicators);
for ( i=3; i<max; i++ ) {
comm = (ompi_communicator_t*)ompi_pointer_array_get_item(&ompi_mpi_communicators,i);
if ( OMPI_COMM_IS_DYNAMIC(comm)) {
objs[j++]=ompi_comm_disconnect_init(comm);
}
}
if ( j != ompi_comm_num_dyncomm+1 ) {
free (objs);
return OMPI_ERROR;
}
ompi_comm_disconnect_waitall (ompi_comm_num_dyncomm, objs);
free (objs);
}
return OMPI_SUCCESS;
}
/**********************************************************************/
/**********************************************************************/
/**********************************************************************/
ompi_comm_disconnect_obj *ompi_comm_disconnect_init ( ompi_communicator_t *comm)
{
ompi_comm_disconnect_obj *obj=NULL;
int ret;
int i;
obj = (ompi_comm_disconnect_obj *) calloc(1,sizeof(ompi_comm_disconnect_obj));
if ( NULL == obj ) {
return NULL;
}
if ( OMPI_COMM_IS_INTER(comm) ) {
obj->size = ompi_comm_remote_size (comm);
}
else {
obj->size = ompi_comm_size (comm);
}
obj->comm = comm;
obj->reqs = (ompi_request_t **) malloc(2*obj->size*sizeof(ompi_request_t *));
if ( NULL == obj->reqs ) {
free (obj);
return NULL;
}
/* initiate all isend_irecvs. We use a dummy buffer stored on
the object, since we are sending zero size messages anyway. */
for ( i=0; i < obj->size; i++ ) {
ret = mca_pml.pml_irecv (&(obj->buf), 0, MPI_INT, i,
OMPI_COMM_BARRIER_TAG, comm,
&(obj->reqs[2*i]));
if ( OMPI_SUCCESS != ret ) {
free (obj->reqs);
free (obj);
return NULL;
}
ret = mca_pml.pml_isend (&(obj->buf), 0, MPI_INT, i,
OMPI_COMM_BARRIER_TAG,
MCA_PML_BASE_SEND_STANDARD,
comm, &(obj->reqs[2*i+1]));
if ( OMPI_SUCCESS != ret ) {
free (obj->reqs);
free (obj);
return NULL;
}
}
/* return handle */
return obj;
}
/**********************************************************************/
/**********************************************************************/
/**********************************************************************/
/* - count how many requests are active
* - generate a request array large enough to hold
all active requests
* - call waitall on the overall request array
* - free the objects
*/
void ompi_comm_disconnect_waitall (int count, ompi_comm_disconnect_obj **objs)
{
ompi_request_t **reqs=NULL;
char *treq=NULL;
int totalcount = 0;
int i;
int ret;
for (i=0; i<count; i++) {
if (NULL == objs[i]) {
printf("Error in comm_disconnect_waitall\n");
return;
}
totalcount += objs[i]->size;
}
reqs = (ompi_request_t **) malloc (2*totalcount*sizeof(ompi_request_t *));
if ( NULL == reqs ) {
printf("ompi_comm_disconnect_waitall: error allocating memory\n");
return;
}
/* generate a single, large array of pending requests */
treq = (char *)reqs;
for (i=0; i<count; i++) {
memcpy (treq, objs[i]->reqs, 2*objs[i]->size * sizeof(ompi_request_t *));
treq += 2*objs[i]->size * sizeof(ompi_request_t *);
}
/* force all non-blocking all-to-alls to finish */
ret = ompi_request_wait_all (2*totalcount, reqs, MPI_STATUSES_IGNORE);
/* Finally, free everything */
for (i=0; i< count; i++ ) {
if (NULL != objs[i]->reqs ) {
free (objs[i]->reqs );
free (objs[i]);
}
}
free (reqs);
/* decrease the counter for dynamic communicators by 'count'.
Attention, this approach now requires, that we are just using
these routines for communicators which have been flagged dynamic */
ompi_comm_num_dyncomm -=count;
return;
}
/**********************************************************************/
/**********************************************************************/
/**********************************************************************/
#define OMPI_COMM_MAXJOBIDS 64
void ompi_comm_mark_dyncomm (ompi_communicator_t *comm)
{
int i, j, numjobids=0;
int size, rsize;
int found;
uint32_t jobids[OMPI_COMM_MAXJOBIDS], thisjobid;
ompi_group_t *grp=NULL;
/* special case for MPI_COMM_NULL */
if ( comm == MPI_COMM_NULL ) {
return;
}
size = ompi_comm_size (comm);
rsize = ompi_comm_remote_size(comm);
/* loop over all processes in local group and count number
of different jobids. */
grp = comm->c_local_group;
for (i=0; i< size; i++) {
thisjobid = ompi_name_server.get_jobid(&(grp->grp_proc_pointers[i]->proc_name));
found = 0;
for ( j=0; j<numjobids; j++) {
if ( thisjobid == jobids[j]) {
found = 1;
break;
}
}
if (!found ) {
jobids[numjobids++] = thisjobid;
}
}
/* if inter-comm, loop over all processes in remote_group
and count number of different jobids */
grp = comm->c_remote_group;
for (i=0; i< rsize; i++) {
thisjobid = ompi_name_server.get_jobid(&(grp->grp_proc_pointers[i]->proc_name));
found = 0;
for ( j=0; j<numjobids; j++) {
if ( thisjobid == jobids[j]) {
found = 1;
break;
}
}
if (!found ) {
jobids[numjobids++] = thisjobid;
}
}
/* if number of joibds larger than one, set the disconnect flag*/
if ( numjobids > 1 ) {
ompi_comm_num_dyncomm++;
OMPI_COMM_SET_DYNAMIC(comm);
}
return;
}