ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Merge pull request #2747 from rhc54/topic/topo

Просмотр исходного кода 
Try a different approach for scalably dealing with hetero clusters
Этот коммит содержится в:
Ralph Castain 2017-01-19 14:22:36 -08:00 коммит произвёл GitHub
родитель cbb28f372a 368684bd63
Коммит 63caeba84d
29 изменённых файлов: 417 добавлений и 974 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

8
orte/mca/ess/base/ess_base_std_orted.c
Просмотреть файл

@ -14,7 +14,7 @@
* Copyright (c) 2011 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011-2013 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2013-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2013-2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -335,7 +335,11 @@ int orte_ess_base_orted_setup(char **hosts)
node->name = strdup(orte_process_info.nodename);
node->index = opal_pointer_array_set_item(orte_node_pool, ORTE_PROC_MY_NAME->vpid, node);
/* point our topology to the one detected locally */
node->topology = opal_hwloc_topology;
node->topology = OBJ_NEW(orte_topology_t);
node->topology->sig = strdup(orte_topo_signature);
node->topology->topo = opal_hwloc_topology;
/* add it to the array of known ones */
opal_pointer_array_add(orte_node_topologies, node->topology);
/* create and store a proc object for us */
proc = OBJ_NEW(orte_proc_t);

28
orte/mca/ess/hnp/ess_hnp_module.c
Просмотреть файл

@ -215,14 +215,6 @@ static int rte_init(void)
goto error;
}
}
/* generate the signature */
orte_topo_signature = opal_hwloc_base_get_topo_signature(opal_hwloc_topology);
if (15 < opal_output_get_verbosity(orte_ess_base_framework.framework_output)) {
opal_output(0, "%s Topology Info:", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
opal_dss.dump(0, opal_hwloc_topology, OPAL_HWLOC_TOPO);
}
/* if we are using xml for output, put an mpirun start tag */
if (orte_xml_output) {
@ -437,12 +429,6 @@ static int rte_init(void)
node->name = strdup(orte_process_info.nodename);
node->index = opal_pointer_array_set_item(orte_node_pool, 0, node);
/* add it to the array of known topologies */
t = OBJ_NEW(orte_topology_t);
t->topo = opal_hwloc_topology;
t->sig = strdup(orte_topo_signature);
opal_pointer_array_add(orte_node_topologies, t);
/* create and store a proc object for us */
proc = OBJ_NEW(orte_proc_t);
proc->name.jobid = ORTE_PROC_MY_NAME->jobid;
@ -521,7 +507,19 @@ static int rte_init(void)
* will have reset our topology. Ensure we always get the right
* one by setting our node topology afterwards
*/
node->topology = opal_hwloc_topology;
/* add it to the array of known topologies */
t = OBJ_NEW(orte_topology_t);
t->topo = opal_hwloc_topology;
/* generate the signature */
orte_topo_signature = opal_hwloc_base_get_topo_signature(opal_hwloc_topology);
t->sig = strdup(orte_topo_signature);
opal_pointer_array_add(orte_node_topologies, t);
node->topology = t;
if (15 < opal_output_get_verbosity(orte_ess_base_framework.framework_output)) {
opal_output(0, "%s Topology Info:", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
opal_dss.dump(0, opal_hwloc_topology, OPAL_HWLOC_TOPO);
}
/* init the hash table, if necessary */
if (NULL == orte_coprocessors) {

5
orte/mca/odls/odls_types.h
Просмотреть файл

@ -12,7 +12,7 @@
* Copyright (c) 2011-2016 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011-2012 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2014-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -86,6 +86,9 @@ typedef uint8_t orte_daemon_cmd_flag_t;
/* for memory profiling */
#define ORTE_DAEMON_GET_MEMPROFILE (orte_daemon_cmd_flag_t) 32
/* request full topology string */
#define ORTE_DAEMON_REPORT_TOPOLOGY_CMD (orte_daemon_cmd_flag_t) 33
/*
* Struct written up the pipe from the child to the parent.
*/

385
orte/mca/plm/base/plm_base_launch_support.c
Просмотреть файл

@ -84,25 +84,25 @@
void orte_plm_base_set_slots(orte_node_t *node)
{
if (0 == strncmp(orte_set_slots, "cores", strlen(orte_set_slots))) {
node->slots = opal_hwloc_base_get_nbobjs_by_type(node->topology,
node->slots = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo,
HWLOC_OBJ_CORE, 0,
OPAL_HWLOC_LOGICAL);
} else if (0 == strncmp(orte_set_slots, "sockets", strlen(orte_set_slots))) {
if (0 == (node->slots = opal_hwloc_base_get_nbobjs_by_type(node->topology,
if (0 == (node->slots = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo,
HWLOC_OBJ_SOCKET, 0,
OPAL_HWLOC_LOGICAL))) {
/* some systems don't report sockets - in this case,
* use numanodes */
node->slots = opal_hwloc_base_get_nbobjs_by_type(node->topology,
node->slots = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo,
HWLOC_OBJ_NODE, 0,
OPAL_HWLOC_LOGICAL);
}
} else if (0 == strncmp(orte_set_slots, "numas", strlen(orte_set_slots))) {
node->slots = opal_hwloc_base_get_nbobjs_by_type(node->topology,
node->slots = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo,
HWLOC_OBJ_NODE, 0,
OPAL_HWLOC_LOGICAL);
} else if (0 == strncmp(orte_set_slots, "hwthreads", strlen(orte_set_slots))) {
node->slots = opal_hwloc_base_get_nbobjs_by_type(node->topology,
node->slots = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo,
HWLOC_OBJ_PU, 0,
OPAL_HWLOC_LOGICAL);
} else {
@ -116,10 +116,8 @@ void orte_plm_base_set_slots(orte_node_t *node)
void orte_plm_base_daemons_reported(int fd, short args, void *cbdata)
{
orte_state_caddy_t *caddy = (orte_state_caddy_t*)cbdata;
hwloc_topology_t t;
orte_job_t *jdata;
orte_topology_t *t;
orte_node_t *node;
orte_proc_t *dmn1;
int i;
/* if we are not launching, then we just assume that all
@ -135,40 +133,6 @@ void orte_plm_base_daemons_reported(int fd, short args, void *cbdata)
node->topology = t;
}
}
} else if (1 < orte_process_info.num_procs) {
/* if we got back topology info from the first node, then we use
* it as the "standard" for all other nodes unless they sent
* back their own topology */
/* find daemon.vpid = 1 */
jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
if (NULL == (dmn1 = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, 1))) {
/* something is wrong */
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
ORTE_FORCED_TERMINATE(ORTE_ERR_NOT_FOUND);
OBJ_RELEASE(caddy);
return;
}
if (NULL == (node = dmn1->node) ||
NULL == (t = node->topology)) {
/* something is wrong */
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
ORTE_FORCED_TERMINATE(ORTE_ERR_NOT_FOUND);
OBJ_RELEASE(caddy);
return;
}
for (i=1; i < orte_node_pool->size; i++) {
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, i))) {
continue;
}
if (NULL == node->topology) {
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s plm:base:setting topo on node %s to that from node %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), node->name, dmn1->node->name));
node->topology = t;
}
}
}
/* if this is an unmanaged allocation, then set the default
@ -856,6 +820,164 @@ void orte_plm_base_registered(int fd, short args, void *cbdata)
static bool orted_failed_launch;
static orte_job_t *jdatorted=NULL;
/* callback for topology reports */
void orte_plm_base_daemon_topology(int status, orte_process_name_t* sender,
opal_buffer_t *buffer,
orte_rml_tag_t tag, void *cbdata)
{
hwloc_topology_t topo;
int rc, idx;
char *sig, *coprocessors, **sns;
orte_proc_t *daemon=NULL;
orte_topology_t *t, *t2;
int i;
uint32_t h;
orte_job_t *jdata;
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s plm:base:daemon_topology for daemon %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(sender)));
/* get the daemon job, if necessary */
if (NULL == jdatorted) {
jdatorted = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
}
if (NULL == (daemon = (orte_proc_t*)opal_pointer_array_get_item(jdatorted->procs, sender->vpid))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
orted_failed_launch = true;
goto CLEANUP;
}
/* unpack the topology signature for this node */
idx=1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &sig, &idx, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
orted_failed_launch = true;
goto CLEANUP;
}
/* find it in the array */
t = NULL;
for (i=0; i < orte_node_topologies->size; i++) {
if (NULL == (t2 = (orte_topology_t*)opal_pointer_array_get_item(orte_node_topologies, i))) {
continue;
}
/* just check the signature */
if (0 == strcmp(sig, t2->sig)) {
t = t2;
break;
}
}
if (NULL == t) {
/* should never happen */
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
orted_failed_launch = true;
goto CLEANUP;
}
/* unpack the topology */
idx=1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &topo, &idx, OPAL_HWLOC_TOPO))) {
ORTE_ERROR_LOG(rc);
orted_failed_launch = true;
goto CLEANUP;
}
/* filter the topology as we'll need it that way later */
opal_hwloc_base_filter_cpus(topo);
/* record the final topology */
t->topo = topo;
/* unpack any coprocessors */
idx=1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &coprocessors, &idx, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
orted_failed_launch = true;
goto CLEANUP;
}
if (NULL != coprocessors) {
/* init the hash table, if necessary */
if (NULL == orte_coprocessors) {
orte_coprocessors = OBJ_NEW(opal_hash_table_t);
opal_hash_table_init(orte_coprocessors, orte_process_info.num_procs);
}
/* separate the serial numbers of the coprocessors
* on this host
*/
sns = opal_argv_split(coprocessors, ',');
for (idx=0; NULL != sns[idx]; idx++) {
/* compute the hash */
OPAL_HASH_STR(sns[idx], h);
/* mark that this coprocessor is hosted by this node */
opal_hash_table_set_value_uint32(orte_coprocessors, h, (void*)&daemon->name.vpid);
}
opal_argv_free(sns);
free(coprocessors);
orte_coprocessors_detected = true;
}
/* see if this daemon is on a coprocessor */
idx=1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &coprocessors, &idx, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
orted_failed_launch = true;
goto CLEANUP;
}
if (NULL != coprocessors) {
if (orte_get_attribute(&daemon->node->attributes, ORTE_NODE_SERIAL_NUMBER, NULL, OPAL_STRING)) {
/* this is not allowed - a coprocessor cannot be host
* to another coprocessor at this time
*/
ORTE_ERROR_LOG(ORTE_ERR_NOT_SUPPORTED);
orted_failed_launch = true;
free(coprocessors);
goto CLEANUP;
}
orte_set_attribute(&daemon->node->attributes, ORTE_NODE_SERIAL_NUMBER, ORTE_ATTR_LOCAL, coprocessors, OPAL_STRING);
free(coprocessors);
orte_coprocessors_detected = true;
}
CLEANUP:
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s plm:base:orted_report_launch %s for daemon %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
orted_failed_launch ? "failed" : "completed",
ORTE_NAME_PRINT(sender)));
if (orted_failed_launch) {
ORTE_ACTIVATE_JOB_STATE(jdatorted, ORTE_JOB_STATE_FAILED_TO_START);
return;
} else {
jdatorted->num_reported++;
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s plm:base:orted_report_launch recvd %d of %d reported daemons",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
jdatorted->num_reported, jdatorted->num_procs));
if (jdatorted->num_procs == jdatorted->num_reported) {
bool dvm = true;
uint32_t key;
void *nptr;
jdatorted->state = ORTE_JOB_STATE_DAEMONS_REPORTED;
/* activate the daemons_reported state for all jobs
* whose daemons were launched
*/
rc = opal_hash_table_get_first_key_uint32(orte_job_data, &key, (void **)&jdata, &nptr);
while (OPAL_SUCCESS == rc) {
if (ORTE_PROC_MY_NAME->jobid != jdata->jobid) {
dvm = false;
if (ORTE_JOB_STATE_DAEMONS_LAUNCHED == jdata->state) {
ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DAEMONS_REPORTED);
}
}
rc = opal_hash_table_get_next_key_uint32(orte_job_data, &key, (void **)&jdata, nptr, &nptr);
}
if (dvm) {
/* must be launching a DVM - activate the state */
ORTE_ACTIVATE_JOB_STATE(jdatorted, ORTE_JOB_STATE_DAEMONS_REPORTED);
}
}
}
}
void orte_plm_base_daemon_callback(int status, orte_process_name_t* sender,
opal_buffer_t *buffer,
orte_rml_tag_t tag, void *cbdata)
@ -867,13 +989,11 @@ void orte_plm_base_daemon_callback(int status, orte_process_name_t* sender,
orte_job_t *jdata;
orte_process_name_t dname;
opal_buffer_t *relay;
char *coprocessors, **sns, *sig;
uint32_t h;
hwloc_topology_t topo;
char *sig;
orte_topology_t *t;
int i;
bool found;
uint8_t tflag;
orte_daemon_cmd_flag_t cmd = ORTE_DAEMON_REPORT_TOPOLOGY_CMD;
/* get the daemon job, if necessary */
if (NULL == jdatorted) {
@ -1024,129 +1144,66 @@ void orte_plm_base_daemon_callback(int status, orte_process_name_t* sender,
free(alias);
}
/* store the local resources for that node */
/* unpack the topology signature for that node */
idx=1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &tflag, &idx, OPAL_UINT8))) {
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &sig, &idx, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
orted_failed_launch = true;
goto CLEANUP;
}
if (1 == tflag) {
idx=1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &sig, &idx, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
orted_failed_launch = true;
goto CLEANUP;
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s RECEIVED TOPOLOGY SIG %s FROM NODE %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), sig, nodename));
/* do we already have this topology from some other node? */
found = false;
for (i=0; i < orte_node_topologies->size; i++) {
if (NULL == (t = (orte_topology_t*)opal_pointer_array_get_item(orte_node_topologies, i))) {
continue;
}
idx=1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &topo, &idx, OPAL_HWLOC_TOPO))) {
ORTE_ERROR_LOG(rc);
orted_failed_launch = true;
goto CLEANUP;
}
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s RECEIVED TOPOLOGY SIG %s FROM NODE %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), sig, nodename));
if (10 < opal_output_get_verbosity(orte_plm_base_framework.framework_output)) {
opal_dss.dump(0, topo, OPAL_HWLOC_TOPO);
}
if (1 == dname.vpid || orte_hetero_nodes) {
/* the user has told us that something is different, so just store it */
/* just check the signature */
if (0 == strcmp(sig, t->sig)) {
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s ADDING TOPOLOGY PER USER REQUEST TO NODE %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), node->name));
t = OBJ_NEW(orte_topology_t);
/* filter the topology as we'll need it that way later */
opal_hwloc_base_filter_cpus(topo);
t->topo = topo;
t->sig = sig;
opal_pointer_array_add(orte_node_topologies, t);
node->topology = topo;
} else {
/* do we already have this topology from some other node? */
found = false;
for (i=0; i < orte_node_topologies->size; i++) {
if (NULL == (t = (orte_topology_t*)opal_pointer_array_get_item(orte_node_topologies, i))) {
continue;
}
/* just check the signature */
if (0 == strcmp(sig, t->sig)) {
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s TOPOLOGY ALREADY RECORDED",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
found = true;
node->topology = t->topo;
hwloc_topology_destroy(topo);
free(sig);
break;
}
}
if (!found) {
/* nope - add it */
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s NEW TOPOLOGY - ADDING",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
t = OBJ_NEW(orte_topology_t);
/* filter the topology as we'll need it that way later */
opal_hwloc_base_filter_cpus(topo);
t->topo = topo;
t->sig = sig;
opal_pointer_array_add(orte_node_topologies, t);
node->topology = topo;
}
"%s TOPOLOGY ALREADY RECORDED",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
found = true;
node->topology = t;
free(sig);
break;
}
}
/* unpack any coprocessors */
idx=1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &coprocessors, &idx, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
orted_failed_launch = true;
goto CLEANUP;
}
if (NULL != coprocessors) {
/* init the hash table, if necessary */
if (NULL == orte_coprocessors) {
orte_coprocessors = OBJ_NEW(opal_hash_table_t);
opal_hash_table_init(orte_coprocessors, orte_process_info.num_procs);
}
/* separate the serial numbers of the coprocessors
* on this host
*/
sns = opal_argv_split(coprocessors, ',');
for (idx=0; NULL != sns[idx]; idx++) {
/* compute the hash */
OPAL_HASH_STR(sns[idx], h);
/* mark that this coprocessor is hosted by this node */
opal_hash_table_set_value_uint32(orte_coprocessors, h, (void*)&node->daemon->name.vpid);
}
opal_argv_free(sns);
free(coprocessors);
orte_coprocessors_detected = true;
}
/* see if this daemon is on a coprocessor */
idx=1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &coprocessors, &idx, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
orted_failed_launch = true;
goto CLEANUP;
}
if (NULL != coprocessors) {
if (orte_get_attribute(&node->attributes, ORTE_NODE_SERIAL_NUMBER, NULL, OPAL_STRING)) {
/* this is not allowed - a coprocessor cannot be host
* to another coprocessor at this time
*/
ORTE_ERROR_LOG(ORTE_ERR_NOT_SUPPORTED);
if (!found) {
/* nope - save the signature and request the complete topology from that node */
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s NEW TOPOLOGY - ADDING",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
t = OBJ_NEW(orte_topology_t);
t->sig = sig;
opal_pointer_array_add(orte_node_topologies, t);
node->topology = t;
/* construct the request */
relay = OBJ_NEW(opal_buffer_t);
if (OPAL_SUCCESS != (rc = opal_dss.pack(relay, &cmd, 1, ORTE_DAEMON_CMD))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(relay);
orted_failed_launch = true;
free(coprocessors);
goto CLEANUP;
}
orte_set_attribute(&node->attributes, ORTE_NODE_SERIAL_NUMBER, ORTE_ATTR_LOCAL, coprocessors, OPAL_STRING);
free(coprocessors);
orte_coprocessors_detected = true;
/* send it */
orte_rml.send_buffer_nb(orte_mgmt_conduit,
sender, relay,
ORTE_RML_TAG_DAEMON,
orte_rml_send_callback, NULL);
/* we will count this node as completed
* when we get the full topology back */
if (NULL != nodename) {
free(nodename);
nodename = NULL;
}
idx = 1;
continue;
}
CLEANUP:
CLEANUP:
OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
"%s plm:base:orted_report_launch %s for daemon %s at contact %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
@ -1208,7 +1265,6 @@ void orte_plm_base_daemon_callback(int status, orte_process_name_t* sender,
ORTE_RML_TAG_DAEMON,
orte_rml_send_callback, NULL);
}
}
void orte_plm_base_daemon_failed(int st, orte_process_name_t* sender,
@ -1251,7 +1307,7 @@ void orte_plm_base_daemon_failed(int st, orte_process_name_t* sender,
daemon->state = ORTE_PROC_STATE_FAILED_TO_START;
daemon->exit_code = status;
finish:
finish:
if (NULL == daemon) {
ORTE_FORCED_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
return;
@ -1331,21 +1387,6 @@ int orte_plm_base_orted_append_basic_args(int *argc, char ***argv,
opal_argv_append(argc, argv, "orte_report_bindings");
opal_argv_append(argc, argv, "1");
}
/* pass our topology signature */
opal_argv_append(argc, argv, "--hnp-topo-sig");
opal_argv_append(argc, argv, orte_topo_signature);
if (orte_hetero_nodes) {
opal_argv_append(argc, argv, "-"OPAL_MCA_CMD_LINE_ID);
opal_argv_append(argc, argv, "orte_hetero_nodes");
opal_argv_append(argc, argv, "1");
}
if (orte_hnp_on_smgmt_node) {
opal_argv_append(argc, argv, "-"OPAL_MCA_CMD_LINE_ID);
opal_argv_append(argc, argv, "orte_hnp_on_smgmt_node");
opal_argv_append(argc, argv, "1");
}
if (orte_map_stddiag_to_stderr) {
opal_argv_append(argc, argv, "-"OPAL_MCA_CMD_LINE_ID);

12
orte/mca/plm/base/plm_base_receive.c
Просмотреть файл

@ -12,7 +12,7 @@
* All rights reserved.
* Copyright (c) 2011 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2014-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* Copyright (c) 2017 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
@ -86,6 +86,10 @@ int orte_plm_base_comm_start(void)
ORTE_RML_TAG_REPORT_REMOTE_LAUNCH,
ORTE_RML_PERSISTENT,
orte_plm_base_daemon_failed, NULL);
orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD,
ORTE_RML_TAG_TOPOLOGY_REPORT,
ORTE_RML_PERSISTENT,
orte_plm_base_daemon_topology, NULL);
}
recv_issued = true;
@ -103,6 +107,12 @@ int orte_plm_base_comm_stop(void)
"%s plm:base:receive stop comm",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_PLM);
if (ORTE_PROC_IS_HNP) {
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_ORTED_CALLBACK);
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_REPORT_REMOTE_LAUNCH);
orte_rml.recv_cancel(ORTE_NAME_WILDCARD, ORTE_RML_TAG_TOPOLOGY_REPORT);
}
recv_issued = false;
return ORTE_SUCCESS;

4
orte/mca/plm/base/plm_private.h
Просмотреть файл

@ -11,6 +11,7 @@
* All rights reserved.
* Copyright (c) 2011-2013 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -79,6 +80,9 @@ ORTE_DECLSPEC void orte_plm_base_daemon_callback(int status, orte_process_name_t
ORTE_DECLSPEC void orte_plm_base_daemon_failed(int status, orte_process_name_t* sender,
opal_buffer_t *buffer,
orte_rml_tag_t tag, void *cbdata);
ORTE_DECLSPEC void orte_plm_base_daemon_topology(int status, orte_process_name_t* sender,
opal_buffer_t *buffer,
orte_rml_tag_t tag, void *cbdata);
ORTE_DECLSPEC int orte_plm_base_create_jobid(orte_job_t *jdata);
ORTE_DECLSPEC int orte_plm_base_set_hnp_name(void);

5
orte/mca/ras/simulator/ras_sim_module.c
Просмотреть файл

@ -3,7 +3,7 @@
* Copyright (c) 2012 Los Alamos National Security, LLC. All rights reserved
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2015 Intel, Inc. All rights reserved
* Copyright (c) 2015-2017 Intel, Inc. All rights reserved.
*
* $COPYRIGHT$
*
@ -266,7 +266,8 @@ static int allocate(orte_job_t *jdata, opal_list_t *nodes)
obj = hwloc_get_root_obj(topo);
node->slots = opal_hwloc_base_get_npus(topo, obj);
}
node->topology = topo;
node->topology = OBJ_NEW(orte_topology_t);
node->topology->topo = topo;
opal_output_verbose(1, orte_ras_base_framework.framework_output,
"Created Node <%10s> [%3d : %3d]",
node->name, node->slots, node->slots_max);

50
orte/mca/rmaps/base/rmaps_base_binding.c
Просмотреть файл

@ -12,7 +12,7 @@
* Copyright (c) 2011-2014 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011-2012 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2013-2016 Intel, Inc. All rights reserved
* Copyright (c) 2013-2017 Intel, Inc. All rights reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
@ -66,7 +66,7 @@ static void reset_usage(orte_node_t *node, orte_jobid_t jobid)
node->name, node->num_procs);
/* start by clearing any existing info */
opal_hwloc_base_clear_usage(node->topology);
opal_hwloc_base_clear_usage(node->topology->topo);
/* cycle thru the procs on the node and record
* their usage in the topology
@ -176,7 +176,7 @@ static int bind_upwards(orte_job_t *jdata,
continue;
}
/* get its index */
if (UINT_MAX == (idx = opal_hwloc_base_get_obj_idx(node->topology, obj, OPAL_HWLOC_AVAILABLE))) {
if (UINT_MAX == (idx = opal_hwloc_base_get_obj_idx(node->topology->topo, obj, OPAL_HWLOC_AVAILABLE))) {
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
return ORTE_ERR_SILENT;
}
@ -184,7 +184,7 @@ static int bind_upwards(orte_job_t *jdata,
data = (opal_hwloc_obj_data_t*)obj->userdata;
data->num_bound++;
/* get the number of cpus under this location */
if (0 == (ncpus = opal_hwloc_base_get_npus(node->topology, obj))) {
if (0 == (ncpus = opal_hwloc_base_get_npus(node->topology->topo, obj))) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-available-cpus", true, node->name);
return ORTE_ERR_SILENT;
}
@ -210,7 +210,7 @@ static int bind_upwards(orte_job_t *jdata,
}
}
/* bind it here */
cpus = opal_hwloc_base_get_available_cpus(node->topology, obj);
cpus = opal_hwloc_base_get_available_cpus(node->topology->topo, obj);
hwloc_bitmap_list_asprintf(&cpu_bitmap, cpus);
orte_set_attribute(&proc->attributes, ORTE_PROC_CPU_BITMAP, ORTE_ATTR_GLOBAL, cpu_bitmap, OPAL_STRING);
/* record the location */
@ -287,7 +287,7 @@ static int bind_downwards(orte_job_t *jdata,
* or if it is some depth below it, so we have to conduct a bit
* of a search. Let hwloc find the min usage one for us.
*/
trg_obj = opal_hwloc_base_find_min_bound_target_under_obj(node->topology, locale,
trg_obj = opal_hwloc_base_find_min_bound_target_under_obj(node->topology->topo, locale,
target, cache_level);
if (NULL == trg_obj) {
/* there aren't any such targets under this object */
@ -310,7 +310,7 @@ static int bind_downwards(orte_job_t *jdata,
}
trg_obj = nxt_obj;
/* get the number of cpus under this location */
ncpus = opal_hwloc_base_get_npus(node->topology, trg_obj);
ncpus = opal_hwloc_base_get_npus(node->topology->topo, trg_obj);
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s GOT %d CPUS",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ncpus);
@ -344,7 +344,7 @@ static int bind_downwards(orte_job_t *jdata,
}
}
/* bind the proc here */
cpus = opal_hwloc_base_get_available_cpus(node->topology, trg_obj);
cpus = opal_hwloc_base_get_available_cpus(node->topology->topo, trg_obj);
hwloc_bitmap_or(totalcpuset, totalcpuset, cpus);
/* track total #cpus */
total_cpus += ncpus;
@ -363,13 +363,13 @@ static int bind_downwards(orte_job_t *jdata,
if (4 < opal_output_get_verbosity(orte_rmaps_base_framework.framework_output)) {
char tmp1[1024], tmp2[1024];
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2str(tmp1, sizeof(tmp1),
node->topology, totalcpuset)) {
node->topology->topo, totalcpuset)) {
opal_output(orte_rmaps_base_framework.framework_output,
"%s PROC %s ON %s IS NOT BOUND",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name), node->name);
} else {
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), node->topology, totalcpuset);
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), node->topology->topo, totalcpuset);
opal_output(orte_rmaps_base_framework.framework_output,
"%s BOUND PROC %s[%s] TO %s: %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
@ -418,7 +418,7 @@ static int bind_in_place(orte_job_t *jdata,
/* if we don't want to launch, then we are just testing the system,
* so ignore questions about support capabilities
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(node->topology);
support = (struct hwloc_topology_support*)hwloc_topology_get_support(node->topology->topo);
/* check if topology supports cpubind - have to be careful here
* as Linux doesn't currently support thread-level binding. This
* may change in the future, though, and it isn't clear how hwloc
@ -461,7 +461,7 @@ static int bind_in_place(orte_job_t *jdata,
* on this node, just silently skip it - we will not bind
*/
if (!OPAL_BINDING_POLICY_IS_SET(map->binding) &&
HWLOC_TYPE_DEPTH_UNKNOWN == hwloc_get_type_depth(node->topology, HWLOC_OBJ_CORE)) {
HWLOC_TYPE_DEPTH_UNKNOWN == hwloc_get_type_depth(node->topology->topo, HWLOC_OBJ_CORE)) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"Unable to bind-to core by default on node %s as no cores detected",
node->name);
@ -490,13 +490,13 @@ static int bind_in_place(orte_job_t *jdata,
return ORTE_ERR_SILENT;
}
/* get the index of this location */
if (UINT_MAX == (idx = opal_hwloc_base_get_obj_idx(node->topology, locale, OPAL_HWLOC_AVAILABLE))) {
if (UINT_MAX == (idx = opal_hwloc_base_get_obj_idx(node->topology->topo, locale, OPAL_HWLOC_AVAILABLE))) {
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
return ORTE_ERR_SILENT;
}
data = (opal_hwloc_obj_data_t*)locale->userdata;
/* get the number of cpus under this location */
if (0 == (ncpus = opal_hwloc_base_get_npus(node->topology, locale))) {
if (0 == (ncpus = opal_hwloc_base_get_npus(node->topology->topo, locale))) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-available-cpus", true, node->name);
return ORTE_ERR_SILENT;
}
@ -511,7 +511,7 @@ static int bind_in_place(orte_job_t *jdata,
found = false;
while (NULL != (sib = sib->next_cousin)) {
data = (opal_hwloc_obj_data_t*)sib->userdata;
ncpus = opal_hwloc_base_get_npus(node->topology, sib);
ncpus = opal_hwloc_base_get_npus(node->topology->topo, sib);
if (data->num_bound < ncpus) {
found = true;
locale = sib;
@ -526,7 +526,7 @@ static int bind_in_place(orte_job_t *jdata,
sib = locale;
while (NULL != (sib = sib->prev_cousin)) {
data = (opal_hwloc_obj_data_t*)sib->userdata;
ncpus = opal_hwloc_base_get_npus(node->topology, sib);
ncpus = opal_hwloc_base_get_npus(node->topology->topo, sib);
if (data->num_bound < ncpus) {
found = true;
locale = sib;
@ -563,7 +563,7 @@ static int bind_in_place(orte_job_t *jdata,
ORTE_NAME_PRINT(&proc->name),
hwloc_obj_type_string(locale->type), idx);
/* bind the proc here */
cpus = opal_hwloc_base_get_available_cpus(node->topology, locale);
cpus = opal_hwloc_base_get_available_cpus(node->topology->topo, locale);
hwloc_bitmap_list_asprintf(&cpu_bitmap, cpus);
orte_set_attribute(&proc->attributes, ORTE_PROC_CPU_BITMAP, ORTE_ATTR_GLOBAL, cpu_bitmap, OPAL_STRING);
/* update the location, in case it changed */
@ -610,7 +610,7 @@ static int bind_to_cpuset(orte_job_t *jdata)
/* if we don't want to launch, then we are just testing the system,
* so ignore questions about support capabilities
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(node->topology);
support = (struct hwloc_topology_support*)hwloc_topology_get_support(node->topology->topo);
/* check if topology supports cpubind - have to be careful here
* as Linux doesn't currently support thread-level binding. This
* may change in the future, though, and it isn't clear how hwloc
@ -643,7 +643,7 @@ static int bind_to_cpuset(orte_job_t *jdata)
}
}
}
root = hwloc_get_root_obj(node->topology);
root = hwloc_get_root_obj(node->topology->topo);
if (NULL == root->userdata) {
/* something went wrong */
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
@ -846,7 +846,7 @@ int orte_rmaps_base_compute_bindings(orte_job_t *jdata)
/* if we don't want to launch, then we are just testing the system,
* so ignore questions about support capabilities
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(node->topology);
support = (struct hwloc_topology_support*)hwloc_topology_get_support(node->topology->topo);
/* check if topology supports cpubind - have to be careful here
* as Linux doesn't currently support thread-level binding. This
* may change in the future, though, and it isn't clear how hwloc
@ -889,7 +889,7 @@ int orte_rmaps_base_compute_bindings(orte_job_t *jdata)
* on this node, just silently skip it - we will not bind
*/
if (!OPAL_BINDING_POLICY_IS_SET(jdata->map->binding) &&
HWLOC_TYPE_DEPTH_UNKNOWN == hwloc_get_type_depth(node->topology, HWLOC_OBJ_CORE)) {
HWLOC_TYPE_DEPTH_UNKNOWN == hwloc_get_type_depth(node->topology->topo, HWLOC_OBJ_CORE)) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"Unable to bind-to core by default on node %s as no cores detected",
node->name);
@ -913,9 +913,9 @@ int orte_rmaps_base_compute_bindings(orte_job_t *jdata)
/* must use a unique function because blasted hwloc
* just doesn't deal with caches very well...sigh
*/
bind_depth = hwloc_get_cache_type_depth(node->topology, clvl, (hwloc_obj_cache_type_t)-1);
bind_depth = hwloc_get_cache_type_depth(node->topology->topo, clvl, (hwloc_obj_cache_type_t)-1);
} else {
bind_depth = hwloc_get_type_depth(node->topology, hwb);
bind_depth = hwloc_get_type_depth(node->topology->topo, hwb);
}
if (0 > bind_depth) {
/* didn't find such an object */
@ -927,9 +927,9 @@ int orte_rmaps_base_compute_bindings(orte_job_t *jdata)
/* must use a unique function because blasted hwloc
* just doesn't deal with caches very well...sigh
*/
map_depth = hwloc_get_cache_type_depth(node->topology, clvm, (hwloc_obj_cache_type_t)-1);
map_depth = hwloc_get_cache_type_depth(node->topology->topo, clvm, (hwloc_obj_cache_type_t)-1);
} else {
map_depth = hwloc_get_type_depth(node->topology, hwm);
map_depth = hwloc_get_type_depth(node->topology->topo, hwm);
}
if (0 > map_depth) {
/* didn't find such an object */

6
orte/mca/rmaps/base/rmaps_base_map_job.c
Просмотреть файл

@ -324,7 +324,7 @@ void orte_rmaps_base_map_job(int fd, short args, void *cbdata)
*/
if (orte_do_not_launch) {
orte_node_t *node;
hwloc_topology_t t0;
orte_topology_t *t0;
int i;
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, 0))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
@ -472,7 +472,7 @@ void orte_rmaps_base_map_job(int fd, short args, void *cbdata)
if (NULL == bd) {
(void)strncpy(tmp1, "UNBOUND", strlen("UNBOUND"));
} else {
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2mapstr(tmp1, sizeof(tmp1), node->topology, bd->cpuset)) {
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2mapstr(tmp1, sizeof(tmp1), node->topology->topo, bd->cpuset)) {
(void)strncpy(tmp1, "UNBOUND", strlen("UNBOUND"));
}
}
@ -497,7 +497,7 @@ void orte_rmaps_base_map_job(int fd, short args, void *cbdata)
}
procbitmap = NULL;
orte_get_attribute(&proc->attributes, ORTE_PROC_CPU_BITMAP, (void**)&procbitmap, OPAL_STRING);
locality = opal_hwloc_base_get_relative_locality(node->topology,
locality = opal_hwloc_base_get_relative_locality(node->topology->topo,
p0bitmap,
procbitmap);
opal_output(orte_clean_output, "\t\t<rank=%s rank=%s locality=%s>",

14
orte/mca/rmaps/base/rmaps_base_ranking.c
Просмотреть файл

@ -10,7 +10,7 @@
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2011 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2014-2015 Intel, Inc. All rights reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -93,7 +93,7 @@ static int rank_span(orte_job_t *jdata,
item = opal_list_get_next(item)) {
node = (orte_node_t*)item;
/* get the number of objects - only consider those we can actually use */
num_objs = opal_hwloc_base_get_nbobjs_by_type(node->topology, target,
num_objs = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo, target,
cache_level, OPAL_HWLOC_AVAILABLE);
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps:rank_span: found %d objects on node %s with %d procs",
@ -104,7 +104,7 @@ static int rank_span(orte_job_t *jdata,
/* for each object */
for (i=0; i < num_objs && cnt < app->num_procs; i++) {
obj = opal_hwloc_base_get_obj_by_type(node->topology, target,
obj = opal_hwloc_base_get_obj_by_type(node->topology->topo, target,
cache_level, i, OPAL_HWLOC_AVAILABLE);
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
@ -206,7 +206,7 @@ static int rank_fill(orte_job_t *jdata,
item = opal_list_get_next(item)) {
node = (orte_node_t*)item;
/* get the number of objects - only consider those we can actually use */
num_objs = opal_hwloc_base_get_nbobjs_by_type(node->topology, target,
num_objs = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo, target,
cache_level, OPAL_HWLOC_AVAILABLE);
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps:rank_fill: found %d objects on node %s with %d procs",
@ -217,7 +217,7 @@ static int rank_fill(orte_job_t *jdata,
/* for each object */
for (i=0; i < num_objs && cnt < app->num_procs; i++) {
obj = opal_hwloc_base_get_obj_by_type(node->topology, target,
obj = opal_hwloc_base_get_obj_by_type(node->topology->topo, target,
cache_level, i, OPAL_HWLOC_AVAILABLE);
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
@ -327,7 +327,7 @@ static int rank_by(orte_job_t *jdata,
item = opal_list_get_next(item)) {
node = (orte_node_t*)item;
/* get the number of objects - only consider those we can actually use */
num_objs = opal_hwloc_base_get_nbobjs_by_type(node->topology, target,
num_objs = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo, target,
cache_level, OPAL_HWLOC_AVAILABLE);
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps:rank_by: found %d objects on node %s with %d procs",
@ -337,7 +337,7 @@ static int rank_by(orte_job_t *jdata,
}
/* collect all the objects */
for (i=0; i < num_objs; i++) {
obj = opal_hwloc_base_get_obj_by_type(node->topology, target,
obj = opal_hwloc_base_get_obj_by_type(node->topology->topo, target,
cache_level, i, OPAL_HWLOC_AVAILABLE);
opal_pointer_array_set_item(&objs, i, obj);
}

30
orte/mca/rmaps/mindist/rmaps_mindist_module.c
Просмотреть файл

@ -12,7 +12,7 @@
* Copyright (c) 2006-2011 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011-2013 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2014-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* Copyright (c) 2014 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
@ -196,7 +196,7 @@ static int mindist_map(orte_job_t *jdata)
item = opal_list_get_next(item)) {
node = (orte_node_t*)item;
if (NULL == node->topology) {
if (NULL == node->topology || NULL == node->topology->topo) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-topology",
true, node->name);
rc = ORTE_ERR_SILENT;
@ -205,7 +205,7 @@ static int mindist_map(orte_job_t *jdata)
/* get the root object as we are not assigning
* locale except at the node level
*/
obj = hwloc_get_root_obj(node->topology);
obj = hwloc_get_root_obj(node->topology->topo);
if (NULL == obj) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-topology",
true, node->name);
@ -215,9 +215,9 @@ static int mindist_map(orte_job_t *jdata)
/* get the number of available pus */
if (opal_hwloc_use_hwthreads_as_cpus) {
total_npus = opal_hwloc_base_get_nbobjs_by_type(node->topology, HWLOC_OBJ_PU, 0, OPAL_HWLOC_AVAILABLE);
total_npus = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo, HWLOC_OBJ_PU, 0, OPAL_HWLOC_AVAILABLE);
} else {
total_npus = opal_hwloc_base_get_nbobjs_by_type(node->topology, HWLOC_OBJ_CORE, 0, OPAL_HWLOC_AVAILABLE);
total_npus = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo, HWLOC_OBJ_CORE, 0, OPAL_HWLOC_AVAILABLE);
}
if (bynode) {
if (total_npus < num_procs_to_assign) {
@ -236,9 +236,9 @@ static int mindist_map(orte_job_t *jdata)
}
/* first we need to fill summary object for root with information about nodes
* so we call opal_hwloc_base_get_nbobjs_by_type */
opal_hwloc_base_get_nbobjs_by_type(node->topology, HWLOC_OBJ_NODE, 0, OPAL_HWLOC_AVAILABLE);
opal_hwloc_base_get_nbobjs_by_type(node->topology->topo, HWLOC_OBJ_NODE, 0, OPAL_HWLOC_AVAILABLE);
OBJ_CONSTRUCT(&numa_list, opal_list_t);
ret = opal_hwloc_get_sorted_numa_list(node->topology, orte_rmaps_base.device, &numa_list);
ret = opal_hwloc_get_sorted_numa_list(node->topology->topo, orte_rmaps_base.device, &numa_list);
if (ret > 1) {
orte_show_help("help-orte-rmaps-md.txt", "orte-rmaps-mindist:several-devices",
true, orte_rmaps_base.device, ret, node->name);
@ -257,11 +257,11 @@ static int mindist_map(orte_job_t *jdata)
required = 0;
OPAL_LIST_FOREACH(numa, &numa_list, opal_rmaps_numa_node_t) {
/* get the hwloc object for this numa */
if (NULL == (obj = opal_hwloc_base_get_obj_by_type(node->topology, HWLOC_OBJ_NODE, 0, numa->index, OPAL_HWLOC_AVAILABLE))) {
if (NULL == (obj = opal_hwloc_base_get_obj_by_type(node->topology->topo, HWLOC_OBJ_NODE, 0, numa->index, OPAL_HWLOC_AVAILABLE))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
npus = opal_hwloc_base_get_npus(node->topology, obj);
npus = opal_hwloc_base_get_npus(node->topology->topo, obj);
if (bynode) {
required = ((num_procs_to_assign-j) > npus) ? (npus) : (num_procs_to_assign-j);
} else {
@ -296,7 +296,7 @@ static int mindist_map(orte_job_t *jdata)
j, node->name);
}
} else {
if (hwloc_get_nbobjs_by_type(node->topology, HWLOC_OBJ_SOCKET) > 1) {
if (hwloc_get_nbobjs_by_type(node->topology->topo, HWLOC_OBJ_SOCKET) > 1) {
/* don't have info about pci locality */
orte_show_help("help-orte-rmaps-md.txt", "orte-rmaps-mindist:no-pci-locality-info",
true, node->name);
@ -355,12 +355,12 @@ static int mindist_map(orte_job_t *jdata)
"mca:rmaps:mindist: second pass assigning %d extra procs to node %s",
(int)num_procs_to_assign, node->name);
OBJ_CONSTRUCT(&numa_list, opal_list_t);
opal_hwloc_get_sorted_numa_list(node->topology, orte_rmaps_base.device, &numa_list);
opal_hwloc_get_sorted_numa_list(node->topology->topo, orte_rmaps_base.device, &numa_list);
if (opal_list_get_size(&numa_list) > 0) {
numa_item = opal_list_get_first(&numa_list);
k = 0;
obj = hwloc_get_obj_by_type(node->topology, HWLOC_OBJ_NODE,((opal_rmaps_numa_node_t*)numa_item)->index);
npus = opal_hwloc_base_get_npus(node->topology, obj);
obj = hwloc_get_obj_by_type(node->topology->topo, HWLOC_OBJ_NODE,((opal_rmaps_numa_node_t*)numa_item)->index);
npus = opal_hwloc_base_get_npus(node->topology->topo, obj);
for (j = 0; j < (int)num_procs_to_assign && nprocs_mapped < (int)app->num_procs; j++) {
if (NULL == (proc = orte_rmaps_base_setup_proc(jdata, node, i))) {
rc = ORTE_ERR_OUT_OF_RESOURCE;
@ -374,8 +374,8 @@ static int mindist_map(orte_job_t *jdata)
if (numa_item == opal_list_get_end(&numa_list)) {
numa_item = opal_list_get_first(&numa_list);
}
obj = hwloc_get_obj_by_type(node->topology, HWLOC_OBJ_NODE,((opal_rmaps_numa_node_t*)numa_item)->index);
npus = opal_hwloc_base_get_npus(node->topology, obj);
obj = hwloc_get_obj_by_type(node->topology->topo, HWLOC_OBJ_NODE,((opal_rmaps_numa_node_t*)numa_item)->index);
npus = opal_hwloc_base_get_npus(node->topology->topo, obj);
k = 0;
}
}

24
orte/mca/rmaps/ppr/rmaps_ppr.c
Просмотреть файл

@ -2,7 +2,7 @@
* Copyright (c) 2011 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2014-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
@ -265,7 +265,7 @@ static int ppr_mapper(orte_job_t *jdata)
item = opal_list_get_next(item)) {
node = (orte_node_t*)item;
/* bozo check */
if (NULL == node->topology) {
if (NULL == node->topology || NULL == node->topology->topo) {
orte_show_help("help-orte-rmaps-ppr.txt", "ppr-topo-missing",
true, node->name);
rc = ORTE_ERR_SILENT;
@ -285,7 +285,7 @@ static int ppr_mapper(orte_job_t *jdata)
* that many procs on this node
*/
if (OPAL_HWLOC_NODE_LEVEL == start) {
obj = hwloc_get_root_obj(node->topology);
obj = hwloc_get_root_obj(node->topology->topo);
for (j=0; j < ppr[start] && nprocs_mapped < total_procs; j++) {
if (NULL == (proc = orte_rmaps_base_setup_proc(jdata, node, idx))) {
rc = ORTE_ERR_OUT_OF_RESOURCE;
@ -296,7 +296,7 @@ static int ppr_mapper(orte_job_t *jdata)
}
} else {
/* get the number of lowest resources on this node */
nobjs = opal_hwloc_base_get_nbobjs_by_type(node->topology,
nobjs = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo,
lowest, cache_level,
OPAL_HWLOC_AVAILABLE);
@ -304,7 +304,7 @@ static int ppr_mapper(orte_job_t *jdata)
* recording the locale of each proc so we know its cpuset
*/
for (i=0; i < nobjs; i++) {
obj = opal_hwloc_base_get_obj_by_type(node->topology,
obj = opal_hwloc_base_get_obj_by_type(node->topology->topo,
lowest, cache_level,
i, OPAL_HWLOC_AVAILABLE);
for (j=0; j < ppr[start] && nprocs_mapped < total_procs; j++) {
@ -486,7 +486,7 @@ static void prune(orte_jobid_t jobid,
}
/* get the number of resources at this level on this node */
nobjs = opal_hwloc_base_get_nbobjs_by_type(node->topology,
nobjs = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo,
lvl, cache_level,
OPAL_HWLOC_AVAILABLE);
@ -494,11 +494,11 @@ static void prune(orte_jobid_t jobid,
* underneath it and check against the limit
*/
for (i=0; i < nobjs; i++) {
obj = opal_hwloc_base_get_obj_by_type(node->topology,
obj = opal_hwloc_base_get_obj_by_type(node->topology->topo,
lvl, cache_level,
i, OPAL_HWLOC_AVAILABLE);
/* get the available cpuset */
avail = opal_hwloc_base_get_available_cpus(node->topology, obj);
avail = opal_hwloc_base_get_available_cpus(node->topology->topo, obj);
/* look at the intersection of this object's cpuset and that
* of each proc in the job/app - if they intersect, then count this proc
@ -518,7 +518,7 @@ static void prune(orte_jobid_t jobid,
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return;
}
cpus = opal_hwloc_base_get_available_cpus(node->topology, locale);
cpus = opal_hwloc_base_get_available_cpus(node->topology->topo, locale);
if (hwloc_bitmap_intersects(avail, cpus)) {
nprocs++;
}
@ -544,7 +544,7 @@ static void prune(orte_jobid_t jobid,
* have only one child, then return this
* object
*/
top = find_split(node->topology, obj);
top = find_split(node->topology->topo, obj);
hwloc_obj_type_snprintf(dang, 64, top, 1);
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps:ppr: SPLIT AT LEVEL %s", dang);
@ -556,7 +556,7 @@ static void prune(orte_jobid_t jobid,
/* find the child with the most procs underneath it */
for (k=0; k < top->arity && limit < nprocs; k++) {
/* get this object's available cpuset */
childcpus = opal_hwloc_base_get_available_cpus(node->topology, top->children[k]);
childcpus = opal_hwloc_base_get_available_cpus(node->topology->topo, top->children[k]);
nunder = 0;
pptr = NULL;
for (n=0; n < node->procs->size; n++) {
@ -572,7 +572,7 @@ static void prune(orte_jobid_t jobid,
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return;
}
cpus = opal_hwloc_base_get_available_cpus(node->topology, locale);
cpus = opal_hwloc_base_get_available_cpus(node->topology->topo, locale);
if (hwloc_bitmap_intersects(childcpus, cpus)) {
nunder++;
if (NULL == pptr) {

6
orte/mca/rmaps/rank_file/rmaps_rank_file.c
Просмотреть файл

@ -14,7 +14,7 @@
* All rights reserved.
* Copyright (c) 2008 Voltaire. All rights reserved
* Copyright (c) 2010 Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2014-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2016 IBM Corporation. All rights reserved.
@ -308,7 +308,7 @@ static int orte_rmaps_rf_map(orte_job_t *jdata)
/* setup the bitmap */
hwloc_cpuset_t bitmap;
char *cpu_bitmap;
if (NULL == node->topology) {
if (NULL == node->topology || NULL == node->topology->topo) {
/* not allowed - for rank-file, we must have
* the topology info
*/
@ -318,7 +318,7 @@ static int orte_rmaps_rf_map(orte_job_t *jdata)
}
bitmap = hwloc_bitmap_alloc();
/* parse the slot_list to find the socket and core */
if (ORTE_SUCCESS != (rc = opal_hwloc_base_slot_list_parse(slots, node->topology, rtype, bitmap))) {
if (ORTE_SUCCESS != (rc = opal_hwloc_base_slot_list_parse(slots, node->topology->topo, rtype, bitmap))) {
ORTE_ERROR_LOG(rc);
hwloc_bitmap_free(bitmap);
goto error;

41
orte/mca/rmaps/round_robin/rmaps_rr_mappers.c
Просмотреть файл

@ -10,7 +10,7 @@
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2009-2013 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2013-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2013-2017 Intel, Inc. All rights reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
@ -76,8 +76,8 @@ int orte_rmaps_rr_byslot(orte_job_t *jdata,
/* get the root object as we are not assigning
* locale here except at the node level
*/
if (NULL != node->topology) {
obj = hwloc_get_root_obj(node->topology);
if (NULL != node->topology && NULL != node->topology->topo) {
obj = hwloc_get_root_obj(node->topology->topo);
}
if (node->slots <= node->slots_inuse) {
opal_output_verbose(2, orte_rmaps_base_framework.framework_output,
@ -143,8 +143,8 @@ int orte_rmaps_rr_byslot(orte_job_t *jdata,
/* get the root object as we are not assigning
* locale except at the node level
*/
if (NULL != node->topology) {
obj = hwloc_get_root_obj(node->topology);
if (NULL != node->topology && NULL != node->topology->topo) {
obj = hwloc_get_root_obj(node->topology->topo);
}
/* add this node to the map - do it only once */
@ -288,8 +288,8 @@ int orte_rmaps_rr_bynode(orte_job_t *jdata,
/* get the root object as we are not assigning
* locale except at the node level
*/
if (NULL != node->topology) {
obj = hwloc_get_root_obj(node->topology);
if (NULL != node->topology && NULL != node->topology->topo) {
obj = hwloc_get_root_obj(node->topology->topo);
}
/* add this node to the map, but only do so once */
if (!ORTE_FLAG_TEST(node, ORTE_NODE_FLAG_MAPPED)) {
@ -402,8 +402,8 @@ int orte_rmaps_rr_bynode(orte_job_t *jdata,
/* get the root object as we are not assigning
* locale except at the node level
*/
if (NULL != node->topology) {
obj = hwloc_get_root_obj(node->topology);
if (NULL != node->topology && NULL != node->topology->topo) {
obj = hwloc_get_root_obj(node->topology->topo);
}
OPAL_OUTPUT_VERBOSE((20, orte_rmaps_base_framework.framework_output,
@ -507,14 +507,14 @@ int orte_rmaps_rr_byobj(orte_job_t *jdata,
do {
add_one = false;
OPAL_LIST_FOREACH(node, node_list, orte_node_t) {
if (NULL == node->topology) {
if (NULL == node->topology || NULL == node->topology->topo) {
orte_show_help("help-orte-rmaps-ppr.txt", "ppr-topo-missing",
true, node->name);
return ORTE_ERR_SILENT;
}
start = 0;
/* get the number of objects of this type on this node */
nobjs = opal_hwloc_base_get_nbobjs_by_type(node->topology, target, cache_level, OPAL_HWLOC_AVAILABLE);
nobjs = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo, target, cache_level, OPAL_HWLOC_AVAILABLE);
if (0 == nobjs) {
continue;
}
@ -564,13 +564,13 @@ int orte_rmaps_rr_byobj(orte_job_t *jdata,
opal_output_verbose(20, orte_rmaps_base_framework.framework_output,
"mca:rmaps:rr: assigning proc to object %d", (i+start) % nobjs);
/* get the hwloc object */
if (NULL == (obj = opal_hwloc_base_get_obj_by_type(node->topology, target, cache_level, (i+start) % nobjs, OPAL_HWLOC_AVAILABLE))) {
if (NULL == (obj = opal_hwloc_base_get_obj_by_type(node->topology->topo, target, cache_level, (i+start) % nobjs, OPAL_HWLOC_AVAILABLE))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
if (orte_rmaps_base.cpus_per_rank > (int)opal_hwloc_base_get_npus(node->topology, obj)) {
if (orte_rmaps_base.cpus_per_rank > (int)opal_hwloc_base_get_npus(node->topology->topo, obj)) {
orte_show_help("help-orte-rmaps-base.txt", "mapping-too-low", true,
orte_rmaps_base.cpus_per_rank, opal_hwloc_base_get_npus(node->topology, obj),
orte_rmaps_base.cpus_per_rank, opal_hwloc_base_get_npus(node->topology->topo, obj),
orte_rmaps_base_print_mapping(orte_rmaps_base.mapping));
return ORTE_ERR_SILENT;
}
@ -663,13 +663,13 @@ static int byobj_span(orte_job_t *jdata,
*/
nobjs = 0;
OPAL_LIST_FOREACH(node, node_list, orte_node_t) {
if (NULL == node->topology) {
if (NULL == node->topology || NULL == node->topology->topo) {
orte_show_help("help-orte-rmaps-ppr.txt", "ppr-topo-missing",
true, node->name);
return ORTE_ERR_SILENT;
}
/* get the number of objects of this type on this node */
nobjs += opal_hwloc_base_get_nbobjs_by_type(node->topology, target, cache_level, OPAL_HWLOC_AVAILABLE);
nobjs += opal_hwloc_base_get_nbobjs_by_type(node->topology->topo, target, cache_level, OPAL_HWLOC_AVAILABLE);
}
if (0 == nobjs) {
@ -708,19 +708,19 @@ static int byobj_span(orte_job_t *jdata,
++(jdata->map->num_nodes);
}
/* get the number of objects of this type on this node */
nobjs = opal_hwloc_base_get_nbobjs_by_type(node->topology, target, cache_level, OPAL_HWLOC_AVAILABLE);
nobjs = opal_hwloc_base_get_nbobjs_by_type(node->topology->topo, target, cache_level, OPAL_HWLOC_AVAILABLE);
opal_output_verbose(2, orte_rmaps_base_framework.framework_output,
"mca:rmaps:rr:byobj: found %d objs on node %s", nobjs, node->name);
/* loop through the number of objects */
for (i=0; i < (int)nobjs && nprocs_mapped < (int)app->num_procs; i++) {
/* get the hwloc object */
if (NULL == (obj = opal_hwloc_base_get_obj_by_type(node->topology, target, cache_level, i, OPAL_HWLOC_AVAILABLE))) {
if (NULL == (obj = opal_hwloc_base_get_obj_by_type(node->topology->topo, target, cache_level, i, OPAL_HWLOC_AVAILABLE))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
return ORTE_ERR_NOT_FOUND;
}
if (orte_rmaps_base.cpus_per_rank > (int)opal_hwloc_base_get_npus(node->topology, obj)) {
if (orte_rmaps_base.cpus_per_rank > (int)opal_hwloc_base_get_npus(node->topology->topo, obj)) {
orte_show_help("help-orte-rmaps-base.txt", "mapping-too-low", true,
orte_rmaps_base.cpus_per_rank, opal_hwloc_base_get_npus(node->topology, obj),
orte_rmaps_base.cpus_per_rank, opal_hwloc_base_get_npus(node->topology->topo, obj),
orte_rmaps_base_print_mapping(orte_rmaps_base.mapping));
return ORTE_ERR_SILENT;
}
@ -759,4 +759,3 @@ static int byobj_span(orte_job_t *jdata,
return ORTE_SUCCESS;
}

18
orte/mca/rmaps/seq/rmaps_seq.c
Просмотреть файл

@ -12,7 +12,7 @@
* Copyright (c) 2006-2011 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2014-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2016 IBM Corporation. All rights reserved.
@ -442,7 +442,7 @@ static int orte_rmaps_seq_map(orte_job_t *jdata)
if (NULL != sq->cpuset) {
hwloc_cpuset_t bitmap;
char *cpu_bitmap;
if (NULL == node->topology) {
if (NULL == node->topology || NULL == node->topology->topo) {
/* not allowed - for sequential cpusets, we must have
* the topology info
*/
@ -460,7 +460,7 @@ static int orte_rmaps_seq_map(orte_job_t *jdata)
/* setup the bitmap */
bitmap = hwloc_bitmap_alloc();
/* parse the slot_list to find the socket and core */
if (ORTE_SUCCESS != (rc = opal_hwloc_base_slot_list_parse(sq->cpuset, node->topology, rtype, bitmap))) {
if (ORTE_SUCCESS != (rc = opal_hwloc_base_slot_list_parse(sq->cpuset, node->topology->topo, rtype, bitmap))) {
ORTE_ERROR_LOG(rc);
hwloc_bitmap_free(bitmap);
goto error;
@ -490,8 +490,8 @@ static int orte_rmaps_seq_map(orte_job_t *jdata)
/* assign the locale - okay for the topo to be null as
* it just means it wasn't returned
*/
if (NULL != node->topology) {
locale = hwloc_get_root_obj(node->topology);
if (NULL != node->topology && NULL != node->topology->topo) {
locale = hwloc_get_root_obj(node->topology->topo);
orte_set_attribute(&proc->attributes, ORTE_PROC_HWLOC_LOCALE,
ORTE_ATTR_LOCAL, locale, OPAL_PTR);
}
@ -531,12 +531,10 @@ static char *orte_getline(FILE *fp)
ret = fgets(input, 1024, fp);
if (NULL != ret) {
input[strlen(input)-1] = '\0'; /* remove newline */
buff = strdup(input);
return buff;
input[strlen(input)-1] = '\0'; /* remove newline */
buff = strdup(input);
return buff;
}
return NULL;
}

35
orte/mca/rmaps/staged/Makefile.am
Просмотреть файл

@ -1,35 +0,0 @@
#
# Copyright (c) 2012 Los Alamos National Security, LLC.
# All rights reserved
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
sources = \
rmaps_staged.c \
rmaps_staged.h \
rmaps_staged_component.c
# Make the output library in this directory, and name it either
# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
# (for static builds).
if MCA_BUILD_orte_rmaps_staged_DSO
component_noinst =
component_install = mca_rmaps_staged.la
else
component_noinst = libmca_rmaps_staged.la
component_install =
endif
mcacomponentdir = $(ortelibdir)
mcacomponent_LTLIBRARIES = $(component_install)
mca_rmaps_staged_la_SOURCES = $(sources)
mca_rmaps_staged_la_LDFLAGS = -module -avoid-version
noinst_LTLIBRARIES = $(component_noinst)
libmca_rmaps_staged_la_SOURCES =$(sources)
libmca_rmaps_staged_la_LDFLAGS = -module -avoid-version

7
orte/mca/rmaps/staged/owner.txt
Просмотреть файл

@ -1,7 +0,0 @@
#
# owner/status file
# owner: institution that is responsible for this package
# status: e.g. active, maintenance, unmaintained
#
owner: INTEL
status: maintenance

436
orte/mca/rmaps/staged/rmaps_staged.c
Просмотреть файл

@ -1,436 +0,0 @@
/*
* Copyright (c) 2012 Los Alamos National Security, LLC.
* All rights reserved
* Copyright (c) 2014 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "orte/types.h"
#include <errno.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif /* HAVE_UNISTD_H */
#include <string.h>
#include "opal/mca/hwloc/base/base.h"
#include "opal/dss/dss.h"
#include "opal/util/argv.h"
#include "opal/util/output.h"
#include "orte/util/show_help.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/state/state.h"
#include "orte/mca/rmaps/base/rmaps_private.h"
#include "orte/mca/rmaps/base/base.h"
#include "rmaps_staged.h"
static int staged_mapper(orte_job_t *jdata);
orte_rmaps_base_module_t orte_rmaps_staged_module = {
staged_mapper
};
static int staged_mapper(orte_job_t *jdata)
{
mca_base_component_t *c=&mca_rmaps_staged_component.base_version;
int i, j, k, rc;
orte_app_context_t *app;
opal_list_t node_list, desired;
orte_std_cntr_t num_slots;
orte_proc_t *proc;
orte_node_t *node, *next;
bool work_to_do = false, first_pass = false;
opal_list_item_t *item, *it2;
char *cptr, **minimap, *hosts, **dash_host;
orte_vpid_t load;
orte_vpid_t max_ppn, *ppn;
/* only use this mapper if it was specified */
if (NULL == jdata->map->req_mapper ||
0 != strcasecmp(jdata->map->req_mapper, c->mca_component_name) ||
ORTE_MAPPING_STAGED != ORTE_GET_MAPPING_POLICY(jdata->map->mapping)) {
/* I wasn't specified */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps:staged: job %s not using staged mapper",
ORTE_JOBID_PRINT(jdata->jobid));
return ORTE_ERR_TAKE_NEXT_OPTION;
}
opal_output_verbose(2, orte_rmaps_base_framework.framework_output,
"%s mca:rmaps:staged: mapping job %s with %d procs",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOBID_PRINT(jdata->jobid), (int)jdata->num_procs);
/* flag that I did the mapping */
if (NULL != jdata->map->last_mapper) {
free(jdata->map->last_mapper);
}
jdata->map->last_mapper = strdup(c->mca_component_name);
/* if there are no nodes in the map, then this is our first
* pass thru this job
*/
if (0 == jdata->map->num_nodes) {
first_pass = true;
}
/* we assume that the app_contexts are in priority order,
* with the highest priority being the first entry in the
* job's app_context array. Loop across the app_contexts
* in order, looking for apps that have not been
* fully mapped
*/
for (i=0; i < jdata->apps->size; i++) {
if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, i))) {
continue;
}
/* has it been fully mapped? */
if (ORTE_APP_STATE_ALL_MAPPED <= app->state) {
continue;
}
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s mca:rmaps:staged: working app %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), app->app);
/* find nodes that meet any constraints provided in the form of
* -hostfile or -host directives
*/
OBJ_CONSTRUCT(&node_list, opal_list_t);
/* get nodes based on a strict interpretation of the location hints */
if (ORTE_SUCCESS != (rc = orte_rmaps_base_get_target_nodes(&node_list, &num_slots, app,
jdata->map->mapping, false, true))) {
/* we were unable to get any nodes that match those
* specified in the app
*/
if (ORTE_ERR_RESOURCE_BUSY == rc) {
/* if the return is "busy", then at least one of the
* specified resources must exist, but no slots are
* currently available. This means there is at least
* a hope of eventually being able to map this app
* within its specified constraints, so continue working
*/
if (orte_soft_locations) {
/* if soft locations were given, then we know that
* none of the nodes in this allocation are available,
* so there is no point in continuing to check the
* remaining apps
*/
while (NULL != (item = opal_list_remove_first(&node_list))) {
OBJ_RELEASE(item);
}
OBJ_DESTRUCT(&node_list);
goto complete;
}
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s mca:rmaps:staged: all nodes for this app are currently busy",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
OBJ_DESTRUCT(&node_list);
continue;
} else {
/* this indicates that there are no nodes that match
* the specified constraints, so there is no hope of
* ever being able to execute this app. This is an
* unrecoverable error - note that a return of
* "silent" means that the function already printed
* an error message, so the error_log will print nothing
*/
ORTE_ERROR_LOG(rc);
return rc;
}
}
/* if a max number of procs/node was given for this
* app, remove all nodes from the list that exceed
* that limit
*/
ppn = &max_ppn;
if (orte_get_attribute(&app->attributes, ORTE_APP_MAX_PPN, (void**)&ppn, OPAL_UINT32)) {
item = opal_list_get_first(&node_list);
while (item != opal_list_get_end(&node_list)) {
it2 = opal_list_get_next(item);
node = (orte_node_t*)item;
if (max_ppn <= node->num_procs) {
opal_list_remove_item(&node_list, item);
OBJ_RELEASE(item);
}
item = it2;
}
}
/* if we have no available nodes, then move on to next app */
if (0 == opal_list_get_size(&node_list)) {
OBJ_DESTRUCT(&node_list);
continue;
}
/* if the app specified locations, soft or not, search the list of nodes
* for those that match the requested locations and move those
* to the desired list so we use them first
*/
if (orte_get_attribute(&app->attributes, ORTE_APP_DASH_HOST, (void**)&hosts, OPAL_STRING)) {
OBJ_CONSTRUCT(&desired, opal_list_t);
dash_host = opal_argv_split(hosts, ',');
free(hosts);
/* no particular order is required */
for (j=0; j < opal_argv_count(dash_host); j++) {
minimap = opal_argv_split(dash_host[j], ',');
for (k=0; k < opal_argv_count(minimap); k++) {
cptr = minimap[k];
for (item = opal_list_get_first(&node_list);
item != opal_list_get_end(&node_list);
item = opal_list_get_next(item)) {
node = (orte_node_t*)item;
if (0 == strcmp(node->name, cptr) ||
(0 == strcmp("localhost", cptr) &&
0 == strcmp(node->name, orte_process_info.nodename))) {
opal_list_remove_item(&node_list, item);
opal_list_append(&desired, item);
opal_output_verbose(10, orte_rmaps_base_framework.framework_output,
"%s mca:rmaps:staged: placing node %s on desired list",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
node->name);
break;
}
}
}
opal_argv_free(minimap);
}
opal_argv_free(dash_host);
/* if no nodes made the transition and the app specified soft
* locations, then we can skip to look at the non-desired list
*/
if (0 == opal_list_get_size(&desired)) {
OBJ_DESTRUCT(&desired);
if (orte_soft_locations) {
goto process;
} else {
/* move on to next app */
continue;
}
}
/* cycle thru the procs for this app and attempt to map them
* to the desired nodes using a load-balancing algo
*/
for (j=0; j < app->procs.size; j++) {
if (NULL == (proc = opal_pointer_array_get_item(&app->procs, j))) {
continue;
}
if (ORTE_PROC_STATE_UNDEF != proc->state) {
/* this proc has already been mapped or executed */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s mca:rmaps:staged: proc %s has already been mapped",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name));
continue;
}
/* flag that there is at least one proc still to
* be executed
*/
work_to_do = true;
/* track number mapped */
jdata->num_mapped++;
/* find the lightest-loaded node on the desired list */
node = NULL;
load = ORTE_VPID_MAX;
for (item = opal_list_get_first(&desired);
item != opal_list_get_end(&desired);
item = opal_list_get_next(item)) {
next = (orte_node_t*)item;
if (next->num_procs < load) {
node = next;
load = next->num_procs;
}
}
/* put the proc there */
proc->node = node;
/* the local rank is the number of procs
* on this node from this job - we don't
* directly track this number, so it must
* be found by looping across the node->procs
* array and counting it each time. For now,
* since we don't use this value in this mode
* of operation, just set it to something arbitrary
*/
proc->local_rank = node->num_procs;
/* the node rank is simply the number of procs
* on the node at this time
*/
proc->node_rank = node->num_procs;
/* track number of procs on node and number of slots used */
node->num_procs++;
node->slots_inuse++;
opal_output_verbose(10, orte_rmaps_base_framework.framework_output,
"%s Proc %s on node %s: slots %d inuse %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name), node->name,
(int)node->slots, (int)node->slots_inuse);
if (node->slots_inuse == node->slots) {
opal_list_remove_item(&desired, &node->super);
OBJ_RELEASE(node);
}
if (0 > (rc = opal_pointer_array_add(node->procs, (void*)proc))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(proc);
return rc;
}
/* retain the proc struct so that we correctly track its release */
OBJ_RETAIN(proc);
proc->state = ORTE_PROC_STATE_INIT;
/* flag the proc as updated so it will be included
* in the next pidmap message
*/
ORTE_FLAG_SET(proc, ORTE_PROC_FLAG_UPDATED);
/* add the node to the map, if needed */
if (!ORTE_FLAG_TEST(node, ORTE_NODE_FLAG_MAPPED)) {
if (ORTE_SUCCESS > (rc = opal_pointer_array_add(jdata->map->nodes, (void*)node))) {
ORTE_ERROR_LOG(rc);
return rc;
}
ORTE_FLAG_SET(node, ORTE_NODE_FLAG_MAPPED);
OBJ_RETAIN(node); /* maintain accounting on object */
jdata->map->num_nodes++;
}
if (0 == opal_list_get_size(&desired)) {
/* nothing more we can do */
break;
}
}
/* clear the list */
while (NULL != (item = opal_list_remove_first(&desired))) {
OBJ_RELEASE(item);
}
OBJ_DESTRUCT(&desired);
}
process:
for (j=0; j < app->procs.size; j++) {
if (NULL == (proc = opal_pointer_array_get_item(&app->procs, j))) {
continue;
}
if (ORTE_PROC_STATE_UNDEF != proc->state) {
/* this proc has already been mapped or executed */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s mca:rmaps:staged: proc %s has already been mapped",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name));
continue;
}
/* find the lightest-loaded node on the node list */
node = NULL;
load = ORTE_VPID_MAX;
for (item = opal_list_get_first(&node_list);
item != opal_list_get_end(&node_list);
item = opal_list_get_next(item)) {
next = (orte_node_t*)item;
if (next->num_procs < load) {
node = next;
load = next->num_procs;
}
}
/* flag that there is at least one proc still to
* be executed
*/
work_to_do = true;
/* track number mapped */
jdata->num_mapped++;
/* map this proc to the first available slot */
OBJ_RETAIN(node); /* maintain accounting on object */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s mca:rmaps:staged: assigning proc %s to node %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name), node->name);
proc->node = node;
/* the local rank is the number of procs
* on this node from this job - we don't
* directly track this number, so it must
* be found by looping across the node->procs
* array and counting it each time. For now,
* since we don't use this value in this mode
* of operation, just set it to something arbitrary
*/
proc->local_rank = node->num_procs;
/* the node rank is simply the number of procs
* on the node at this time
*/
proc->node_rank = node->num_procs;
/* track number of procs on node and number of slots used */
node->num_procs++;
node->slots_inuse++;
opal_output_verbose(10, orte_rmaps_base_framework.framework_output,
"%s Proc %s on node %s: slots %d inuse %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name), node->name,
(int)node->slots, (int)node->slots_inuse);
if (node->slots_inuse == node->slots) {
opal_list_remove_item(&node_list, &node->super);
OBJ_RELEASE(node);
}
if (0 > (rc = opal_pointer_array_add(node->procs, (void*)proc))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(proc);
return rc;
}
/* retain the proc struct so that we correctly track its release */
OBJ_RETAIN(proc);
proc->state = ORTE_PROC_STATE_INIT;
/* flag the proc as updated so it will be included
* in the next pidmap message
*/
ORTE_FLAG_SET(proc, ORTE_PROC_FLAG_UPDATED);
/* add the node to the map, if needed */
if (!ORTE_FLAG_TEST(node, ORTE_NODE_FLAG_MAPPED)) {
if (ORTE_SUCCESS > (rc = opal_pointer_array_add(jdata->map->nodes, (void*)node))) {
ORTE_ERROR_LOG(rc);
return rc;
}
ORTE_FLAG_SET(node, ORTE_NODE_FLAG_MAPPED);
OBJ_RETAIN(node); /* maintain accounting on object */
jdata->map->num_nodes++;
}
if (0 == opal_list_get_size(&node_list)) {
/* nothing more we can do */
break;
}
}
/* clear the list */
while (NULL != (item = opal_list_remove_first(&node_list))) {
OBJ_RELEASE(item);
}
OBJ_DESTRUCT(&node_list);
}
complete:
/* if there isn't at least one proc that can be launched,
* then indicate that we don't need to proceed with the
* launch sequence
*/
if (!work_to_do) {
return ORTE_ERR_RESOURCE_BUSY;
}
/* flag that the job was updated so it will be
* included in the pidmap message
*/
ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_UPDATED);
/* if we successfully mapped ALL procs in the first pass,
* then this job is capable of supporting MPI procs
*/
if (first_pass && jdata->num_mapped == jdata->num_procs) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s mca:rmaps:staged: job %s is MPI-capable",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOBID_PRINT(jdata->jobid));
ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_GANG_LAUNCHED);
}
return ORTE_SUCCESS;
}

25
orte/mca/rmaps/staged/rmaps_staged.h
Просмотреть файл

@ -1,25 +0,0 @@
/*
* Copyright (c) 2012 Los Alamos National Security, LLC.
* All rights reserved
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#ifndef ORTE_RMAPS_STAGED_H
#define ORTE_RMAPS_STAGED_H
#include "orte_config.h"
#include "orte/mca/rmaps/rmaps.h"
BEGIN_C_DECLS
ORTE_MODULE_DECLSPEC extern orte_rmaps_base_component_t mca_rmaps_staged_component;
extern orte_rmaps_base_module_t orte_rmaps_staged_module;
END_C_DECLS
#endif

73
orte/mca/rmaps/staged/rmaps_staged_component.c
Просмотреть файл

@ -1,73 +0,0 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2012-2015 Los Alamos National Security, LLC. All rights
* reserved
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "opal/mca/base/base.h"
#include "orte/util/show_help.h"
#include "orte/mca/rmaps/base/base.h"
#include "rmaps_staged.h"
/*
* Local functions
*/
static int orte_rmaps_staged_open(void);
static int orte_rmaps_staged_close(void);
static int orte_rmaps_staged_query(mca_base_module_t **module, int *priority);
orte_rmaps_base_component_t mca_rmaps_staged_component = {
.base_version = {
ORTE_RMAPS_BASE_VERSION_2_0_0,
.mca_component_name = "staged",
MCA_BASE_MAKE_VERSION(component, ORTE_MAJOR_VERSION, ORTE_MINOR_VERSION,
ORTE_RELEASE_VERSION),
.mca_open_component = orte_rmaps_staged_open,
.mca_close_component = orte_rmaps_staged_close,
.mca_query_component = orte_rmaps_staged_query,
},
.base_data = {
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
},
};
/**
* component open/close/init function
*/
static int orte_rmaps_staged_open(void)
{
return ORTE_SUCCESS;
}
static int orte_rmaps_staged_query(mca_base_module_t **module, int *priority)
{
*priority = 5;
*module = (mca_base_module_t *)&orte_rmaps_staged_module;
return ORTE_SUCCESS;
}
/**
* Close all subsystems.
*/
static int orte_rmaps_staged_close(void)
{
return ORTE_SUCCESS;
}

7
orte/mca/rml/rml_types.h
Просмотреть файл

@ -12,7 +12,7 @@
* Copyright (c) 2007-2012 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2009-2016 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2014-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -169,6 +169,9 @@ BEGIN_C_DECLS
/* memory profile */
#define ORTE_RML_TAG_MEMPROFILE 61
/* topology report */
#define ORTE_RML_TAG_TOPOLOGY_REPORT 62
#define ORTE_RML_TAG_MAX 100
@ -178,7 +181,7 @@ BEGIN_C_DECLS
/*** length of the tag. change this when type of orte_rml_tag_t is changed ***/
/*** max valu in unit32_t is 0xFFFF_FFFF when converted to char this is 8 **
#define ORTE_RML_TAG_T_CHAR_LEN 8
#define ORTE_RML_TAG_T_SPRINT "%8x" */
#define ORTE_RML_TAG_T_SPRINT "%8x" */
/**
* Message matching tag

11
orte/mca/schizo/ompi/schizo_ompi.c
Просмотреть файл

@ -13,7 +13,7 @@
* All rights reserved.
* Copyright (c) 2009-2016 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011 Oak Ridge National Labs. All rights reserved.
* Copyright (c) 2013-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2013-2017 Intel, Inc. All rights reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
@ -114,11 +114,6 @@ static opal_cmd_line_init_t cmd_line_init[] = {
&orte_cmd_options.hnp, OPAL_CMD_LINE_TYPE_STRING,
"Specify the URI of the HNP, or the name of the file (specified as file:filename) that contains that info" },
/* hetero apps */
{ "orte_hetero_apps", '\0', NULL, "hetero-apps", 0,
NULL, OPAL_CMD_LINE_TYPE_BOOL,
"Indicates that multiple app_contexts are being provided that are a mix of 32/64 bit binaries" },
/* select XML output */
{ "orte_xml_output", '\0', "xml", "xml", 0,
NULL, OPAL_CMD_LINE_TYPE_BOOL,
@ -422,10 +417,6 @@ static opal_cmd_line_init_t cmd_line_init[] = {
&orte_cmd_options.continuous, OPAL_CMD_LINE_TYPE_BOOL,
"Job is to run until explicitly terminated" },
{ "orte_hetero_nodes", '\0', NULL, "hetero-nodes", 0,
NULL, OPAL_CMD_LINE_TYPE_BOOL,
"Nodes in cluster may differ in topology, so send the topology back from each node [Default = false]" },
#if OPAL_ENABLE_CRDEBUG == 1
{ "opal_cr_enable_crdebug", '\0', "crdebug", "crdebug", 0,
NULL, OPAL_CMD_LINE_TYPE_BOOL,

4
orte/mca/state/novm/state_novm.c
Просмотреть файл

@ -1,7 +1,7 @@
/*
* Copyright (c) 2011-2012 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2014-2015 Intel, Inc. All rights reserved
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -222,7 +222,7 @@ static void allocation_complete(int fd, short args, void *cbdata)
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, i))) {
continue;
}
node->topology = t->topo;
node->topology = t;
}
/* move to the map stage */

44
orte/orted/orted_comm.c
Просмотреть файл

@ -14,7 +14,7 @@
* reserved.
* Copyright (c) 2009 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2010-2011 Oak Ridge National Labs. All rights reserved.
* Copyright (c) 2014-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
* Copyright (c) 2016 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
@ -121,6 +121,7 @@ void orte_daemon_recv(int status, orte_process_name_t* sender,
float pss;
opal_pstats_t pstat;
char *rtmod;
char *coprocessors;
/* unpack the command */
n = 1;
@ -548,6 +549,47 @@ void orte_daemon_recv(int status, orte_process_name_t* sender,
}
break;
/**** REPORT TOPOLOGY COMMAND ****/
case ORTE_DAEMON_REPORT_TOPOLOGY_CMD:
answer = OBJ_NEW(opal_buffer_t);
/* pack the topology signature */
if (ORTE_SUCCESS != (ret = opal_dss.pack(answer, &orte_topo_signature, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(answer);
goto CLEANUP;
}
/* pack the topology */
if (ORTE_SUCCESS != (ret = opal_dss.pack(answer, &opal_hwloc_topology, 1, OPAL_HWLOC_TOPO))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(answer);
goto CLEANUP;
}
/* detect and add any coprocessors */
coprocessors = opal_hwloc_base_find_coprocessors(opal_hwloc_topology);
if (ORTE_SUCCESS != (ret = opal_dss.pack(answer, &coprocessors, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
}
if (NULL != coprocessors) {
free(coprocessors);
}
/* see if I am on a coprocessor */
coprocessors = opal_hwloc_base_check_on_coprocessor();
if (ORTE_SUCCESS != (ret = opal_dss.pack(answer, &coprocessors, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
}
if (NULL!= coprocessors) {
free(coprocessors);
}
/* send the data */
if (0 > (ret = orte_rml.send_buffer_nb(orte_mgmt_conduit,
sender, answer, ORTE_RML_TAG_TOPOLOGY_REPORT,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(ret);
OBJ_RELEASE(answer);
}
break;
/**** CONTACT QUERY COMMAND ****/
case ORTE_DAEMON_CONTACT_QUERY_CMD:
if (orte_debug_daemons_flag) {

50
orte/orted/orted_main.c
Просмотреть файл

@ -127,9 +127,7 @@ static struct {
int singleton_died_pipe;
bool abort;
bool tree_spawn;
char *hnp_topo_sig;
bool test_suicide;
bool hnp_on_smgmt_node;
} orted_globals;
/*
@ -209,18 +207,6 @@ opal_cmd_line_init_t orte_cmd_line_opts[] = {
NULL, OPAL_CMD_LINE_TYPE_STRING,
"Regular expression defining nodes in system" },
{ "orte_hetero_nodes", '\0', NULL, "hetero-nodes", 0,
NULL, OPAL_CMD_LINE_TYPE_BOOL,
"Nodes in cluster may differ in topology, so send the topology back from each node [Default = false]" },
{ NULL, '\0', NULL, "hnp-topo-sig", 1,
&orted_globals.hnp_topo_sig, OPAL_CMD_LINE_TYPE_STRING,
"Topology signature of HNP" },
{ "orte_hnp_on_smgmt_node", '\0', NULL, "hnp-on-smgmt-node", 0,
&orted_globals.hnp_on_smgmt_node, OPAL_CMD_LINE_TYPE_BOOL,
"Mpirun is executing on a system mgmt node whose topology is different from the compute nodes [Default = false]" },
/* End of list */
{ NULL, '\0', NULL, NULL, 0,
NULL, OPAL_CMD_LINE_TYPE_NULL, NULL }
@ -237,8 +223,6 @@ int orte_daemon(int argc, char *argv[])
#if OPAL_ENABLE_FT_CR == 1
char *tmp_env_var = NULL;
#endif
char *coprocessors;
uint8_t tflag;
/* initialize the globals */
memset(&orted_globals, 0, sizeof(orted_globals));
@ -769,39 +753,11 @@ int orte_daemon(int argc, char *argv[])
opal_argv_free(aliases);
}
/* add the local topology, if different from the HNP's or user directed us to,
* but always if we are the first daemon to ensure we get a compute node */
if (1 == ORTE_PROC_MY_NAME->vpid || orte_hetero_nodes ||
(!orted_globals.hnp_on_smgmt_node && 0 != strcmp(orte_topo_signature, orted_globals.hnp_topo_sig))) {
tflag = 1;
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &tflag, 1, OPAL_UINT8))) {
ORTE_ERROR_LOG(ret);
}
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &orte_topo_signature, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
}
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &opal_hwloc_topology, 1, OPAL_HWLOC_TOPO))) {
ORTE_ERROR_LOG(ret);
}
} else {
tflag = 0;
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &tflag, 1, OPAL_UINT8))) {
ORTE_ERROR_LOG(ret);
}
}
/* detect and add any coprocessors */
coprocessors = opal_hwloc_base_find_coprocessors(opal_hwloc_topology);
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &coprocessors, 1, OPAL_STRING))) {
/* always send back our topology signature - this is a small string
* and won't hurt anything */
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &orte_topo_signature, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
}
/* see if I am on a coprocessor */
coprocessors = opal_hwloc_base_check_on_coprocessor();
if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &coprocessors, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(ret);
}
if (NULL!= coprocessors) {
free(coprocessors);
}
/* send to the HNP's callback - will be routed if routes are available */
if (0 > (ret = orte_rml.send_buffer_nb(orte_coll_conduit,

13
orte/runtime/data_type_support/orte_dt_print_fns.c
Просмотреть файл

@ -13,7 +13,7 @@
* Copyright (c) 2011-2015 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2011-2013 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2013-2016 Intel, Inc. All rights reserved.
* Copyright (c) 2013-2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -476,13 +476,13 @@ int orte_dt_print_proc(char **output, char *prefix, orte_proc_t *src, opal_data_
char *str=NULL, *cpu_bitmap=NULL;
if (orte_get_attribute(&src->attributes, ORTE_PROC_CPU_BITMAP, (void**)&cpu_bitmap, OPAL_STRING) &&
NULL != src->node->topology) {
NULL != src->node->topology && NULL != src->node->topology->topo) {
mycpus = hwloc_bitmap_alloc();
hwloc_bitmap_list_sscanf(mycpus, cpu_bitmap);
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2str(tmp1, sizeof(tmp1), src->node->topology, mycpus)) {
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2str(tmp1, sizeof(tmp1), src->node->topology->topo, mycpus)) {
str = strdup("UNBOUND");
} else {
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), src->node->topology, mycpus);
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), src->node->topology->topo, mycpus);
asprintf(&str, "%s:%s", tmp1, tmp2);
}
hwloc_bitmap_free(mycpus);
@ -517,7 +517,7 @@ int orte_dt_print_proc(char **output, char *prefix, orte_proc_t *src, opal_data_
if (orte_get_attribute(&src->attributes, ORTE_PROC_HWLOC_LOCALE, (void**)&loc, OPAL_PTR)) {
if (NULL != loc) {
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2mapstr(locale, sizeof(locale), src->node->topology, loc->cpuset)) {
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2mapstr(locale, sizeof(locale), src->node->topology->topo, loc->cpuset)) {
strcpy(locale, "NODE");
}
} else {
@ -528,7 +528,7 @@ int orte_dt_print_proc(char **output, char *prefix, orte_proc_t *src, opal_data_
}
if (orte_get_attribute(&src->attributes, ORTE_PROC_HWLOC_BOUND, (void**)&bd, OPAL_PTR)) {
if (NULL != bd) {
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2mapstr(bind, sizeof(bind), src->node->topology, bd->cpuset)) {
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2mapstr(bind, sizeof(bind), src->node->topology->topo, bd->cpuset)) {
strcpy(bind, "UNBOUND");
}
} else {
@ -869,4 +869,3 @@ int orte_dt_print_sig(char **output, char *prefix, orte_grpcomm_signature_t *src
*output = tmp;
return ORTE_SUCCESS;
}

3
orte/runtime/orte_globals.c
Просмотреть файл

@ -90,9 +90,6 @@ int orte_hostname_cutoff = 1000;
int orted_debug_failure = -1;
int orted_debug_failure_delay = -1;
bool orte_hetero_apps = false;
bool orte_hetero_nodes = false;
bool orte_hnp_on_smgmt_node = false;
bool orte_never_launched = false;
bool orte_devel_level_output = false;
bool orte_display_topo_with_map = false;

24
orte/runtime/orte_globals.h
Просмотреть файл

@ -215,6 +215,15 @@ struct orte_proc_t;
struct orte_job_map_t;
/************/
/* define an object for storing node topologies */
typedef struct {
opal_object_t super;
hwloc_topology_t topo;
char *sig;
} orte_topology_t;
ORTE_DECLSPEC OBJ_CLASS_DECLARATION(orte_topology_t);
/**
* Information about a specific application to be launched in the RTE.
*/
@ -292,7 +301,7 @@ typedef struct {
may want to allow up to four processes but no more. */
orte_std_cntr_t slots_max;
/* system topology for this node */
hwloc_topology_t topology;
orte_topology_t *topology;
/* flags */
orte_node_flags_t flags;
/* list of orte_attribute_t */
@ -405,14 +414,6 @@ struct orte_proc_t {
typedef struct orte_proc_t orte_proc_t;
ORTE_DECLSPEC OBJ_CLASS_DECLARATION(orte_proc_t);
/* define an object for storing node topologies */
typedef struct {
opal_object_t super;
hwloc_topology_t topo;
char *sig;
} orte_topology_t;
ORTE_DECLSPEC OBJ_CLASS_DECLARATION(orte_topology_t);
/**
* Get a job data object
* We cannot just reference a job data object with its jobid as
@ -472,11 +473,6 @@ ORTE_DECLSPEC extern int orte_hostname_cutoff;
ORTE_DECLSPEC extern int orted_debug_failure;
ORTE_DECLSPEC extern int orted_debug_failure_delay;
/* homegeneity flags */
ORTE_DECLSPEC extern bool orte_hetero_apps;
ORTE_DECLSPEC extern bool orte_hetero_nodes;
ORTE_DECLSPEC extern bool orte_hnp_on_smgmt_node;
ORTE_DECLSPEC extern bool orte_never_launched;
ORTE_DECLSPEC extern bool orte_devel_level_output;
ORTE_DECLSPEC extern bool orte_display_topo_with_map;

23
orte/runtime/orte_mca_params.c
Просмотреть файл

@ -510,21 +510,6 @@ int orte_register_params(void)
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_READONLY,
&orte_show_resolved_nodenames);
#if 0
/* XXX -- option doesn't appear to do anything */
mca_base_param_reg_int_name("orte", "hetero_apps",
"Indicates that multiple app_contexts are being provided that are a mix of 32/64 bit binaries (default: false)",
false, false, (int) false, &value);
orte_hetero_apps = OPAL_INT_TO_BOOL(value);
#endif
orte_hetero_nodes = false;
(void) mca_base_var_register ("orte", "orte", NULL, "hetero_nodes",
"Nodes in cluster may differ in topology, so send the topology back from each node [Default = false]",
MCA_BASE_VAR_TYPE_BOOL, NULL, 0, 0,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_READONLY,
&orte_hetero_nodes);
/* allow specification of the launch agent */
orte_launch_agent = "orted";
(void) mca_base_var_register ("orte", "orte", NULL, "launch_agent",
@ -767,13 +752,5 @@ int orte_register_params(void)
MCA_BASE_VAR_TYPE_STRING, NULL, 0, 0, OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY, &orte_mgmt_transport);
orte_hnp_on_smgmt_node = false;
(void) mca_base_var_register ("orte", "orte", NULL, "hnp_on_smgmt_node",
"Mpirun is executing on a system mgmt node whose topology is different from the compute nodes [Default = false]",
MCA_BASE_VAR_TYPE_BOOL, NULL, 0, 0,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_READONLY,
&orte_hnp_on_smgmt_node);
return ORTE_SUCCESS;
}