int opal_hwloc_unpack(opal_buffer_t *buffer, void *dest,
int32_t *num_vals,
opal_data_type_t type)
{
/* NOTE: hwloc defines topology_t as a pointer to a struct! */
hwloc_topology_t t, *tarray = (hwloc_topology_t*)dest;
int rc=OPAL_SUCCESS, i, cnt, j;
char *xmlbuffer;
struct hwloc_topology_support *support;
for (i=0, j=0; i < *num_vals; i++) {
/* unpack the xml string */
cnt=1;
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &xmlbuffer, &cnt, OPAL_STRING))) {
goto cleanup;
}
/* convert the xml */
if (0 != hwloc_topology_init(&t)) {
rc = OPAL_ERROR;
free(xmlbuffer);
goto cleanup;
}
if (0 != hwloc_topology_set_xmlbuffer(t, xmlbuffer, strlen(xmlbuffer))) {
rc = OPAL_ERROR;
free(xmlbuffer);
hwloc_topology_destroy(t);
goto cleanup;
}
free(xmlbuffer);
/* since we are loading this from an external source, we have to
* explicitly set a flag so hwloc sets things up correctly
*/
if (0 != hwloc_topology_set_flags(t, (HWLOC_TOPOLOGY_FLAG_IS_THISSYSTEM |
HWLOC_TOPOLOGY_FLAG_WHOLE_SYSTEM |
HWLOC_TOPOLOGY_FLAG_IO_DEVICES))) {
rc = OPAL_ERROR;
hwloc_topology_destroy(t);
goto cleanup;
}
/* now load the topology */
if (0 != hwloc_topology_load(t)) {
rc = OPAL_ERROR;
hwloc_topology_destroy(t);
goto cleanup;
}
/* get the available support - hwloc unfortunately does
* not include this info in its xml import!
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(t);
cnt = sizeof(struct hwloc_topology_discovery_support);
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, support->discovery, &cnt, OPAL_BYTE))) {
goto cleanup;
}
cnt = sizeof(struct hwloc_topology_cpubind_support);
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, support->cpubind, &cnt, OPAL_BYTE))) {
goto cleanup;
}
cnt = sizeof(struct hwloc_topology_membind_support);
if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, support->membind, &cnt, OPAL_BYTE))) {
goto cleanup;
}
/* filter the cpus thru any default cpu set */
if (OPAL_SUCCESS != (rc = opal_hwloc_base_filter_cpus(t))) {
goto cleanup;
}
/* pass it back */
tarray[i] = t;
/* track the number added */
j++;
}
cleanup:
*num_vals = j;
return rc;
}
static int rte_init(void)
{
int ret;
char *error = NULL;
char *envar, *ev1, *ev2;
uint64_t unique_key[2];
char *string_key;
char *rmluri;
opal_value_t *kv;
char *val;
int u32, *u32ptr;
uint16_t u16, *u16ptr;
char **peers=NULL, *mycpuset, **cpusets=NULL;
opal_process_name_t name;
size_t i;
/* run the prolog */
if (ORTE_SUCCESS != (ret = orte_ess_base_std_prolog())) {
error = "orte_ess_base_std_prolog";
goto error;
}
/* get an async event base - we use the opal_async one so
* we don't startup extra threads if not needed */
orte_event_base = opal_progress_thread_init(NULL);
progress_thread_running = true;
/* open and setup pmix */
if (OPAL_SUCCESS != (ret = mca_base_framework_open(&opal_pmix_base_framework, 0))) {
ORTE_ERROR_LOG(ret);
/* we cannot run */
error = "pmix init";
goto error;
}
if (OPAL_SUCCESS != (ret = opal_pmix_base_select())) {
/* we cannot run */
error = "pmix init";
goto error;
}
/* set the event base */
opal_pmix_base_set_evbase(orte_event_base);
/* initialize the selected module */
if (!opal_pmix.initialized() && (OPAL_SUCCESS != (ret = opal_pmix.init()))) {
/* we cannot run */
error = "pmix init";
goto error;
}
u32ptr = &u32;
u16ptr = &u16;
/**** THE FOLLOWING ARE REQUIRED VALUES ***/
/* pmix.init set our process name down in the OPAL layer,
* so carry it forward here */
ORTE_PROC_MY_NAME->jobid = OPAL_PROC_MY_NAME.jobid;
ORTE_PROC_MY_NAME->vpid = OPAL_PROC_MY_NAME.vpid;
/* get our local rank from PMI */
OPAL_MODEX_RECV_VALUE(ret, OPAL_PMIX_LOCAL_RANK,
ORTE_PROC_MY_NAME, &u16ptr, OPAL_UINT16);
if (OPAL_SUCCESS != ret) {
error = "getting local rank";
goto error;
}
orte_process_info.my_local_rank = u16;
/* get our node rank from PMI */
OPAL_MODEX_RECV_VALUE(ret, OPAL_PMIX_NODE_RANK,
ORTE_PROC_MY_NAME, &u16ptr, OPAL_UINT16);
if (OPAL_SUCCESS != ret) {
error = "getting node rank";
goto error;
}
orte_process_info.my_node_rank = u16;
/* get max procs */
OPAL_MODEX_RECV_VALUE(ret, OPAL_PMIX_MAX_PROCS,
ORTE_PROC_MY_NAME, &u32ptr, OPAL_UINT32);
if (OPAL_SUCCESS != ret) {
error = "getting max procs";
goto error;
}
orte_process_info.max_procs = u32;
/* get job size */
OPAL_MODEX_RECV_VALUE(ret, OPAL_PMIX_JOB_SIZE,
ORTE_PROC_MY_NAME, &u32ptr, OPAL_UINT32);
if (OPAL_SUCCESS != ret) {
error = "getting job size";
goto error;
}
orte_process_info.num_procs = u32;
/* push into the environ for pickup in MPI layer for
* MPI-3 required info key
*/
if (NULL == getenv(OPAL_MCA_PREFIX"orte_ess_num_procs")) {
asprintf(&ev1, OPAL_MCA_PREFIX"orte_ess_num_procs=%d", orte_process_info.num_procs);
putenv(ev1);
added_num_procs = true;
}
if (NULL == getenv("OMPI_APP_CTX_NUM_PROCS")) {
asprintf(&ev2, "OMPI_APP_CTX_NUM_PROCS=%d", orte_process_info.num_procs);
putenv(ev2);
added_app_ctx = true;
}
/* get our app number from PMI - ok if not found */
OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_APPNUM,
ORTE_PROC_MY_NAME, &u32ptr, OPAL_UINT32);
if (OPAL_SUCCESS == ret) {
orte_process_info.app_num = u32;
} else {
orte_process_info.app_num = 0;
}
/* get the number of local peers - required for wireup of
* shared memory BTL */
OPAL_MODEX_RECV_VALUE(ret, OPAL_PMIX_LOCAL_SIZE,
ORTE_PROC_MY_NAME, &u32ptr, OPAL_UINT32);
if (OPAL_SUCCESS == ret) {
orte_process_info.num_local_peers = u32 - 1; // want number besides ourselves
} else {
orte_process_info.num_local_peers = 0;
}
/* setup transport keys in case the MPI layer needs them -
* we can use the jobfam and stepid as unique keys
* because they are unique values assigned by the RM
*/
if (NULL == getenv(OPAL_MCA_PREFIX"orte_precondition_transports")) {
unique_key[0] = ORTE_JOB_FAMILY(ORTE_PROC_MY_NAME->jobid);
unique_key[1] = ORTE_LOCAL_JOBID(ORTE_PROC_MY_NAME->jobid);
if (NULL == (string_key = orte_pre_condition_transports_print(unique_key))) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
opal_output_verbose(2, orte_ess_base_framework.framework_output,
"%s transport key %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), string_key);
asprintf(&envar, OPAL_MCA_PREFIX"orte_precondition_transports=%s", string_key);
putenv(envar);
added_transport_keys = true;
/* cannot free the envar as that messes up our environ */
free(string_key);
}
/* retrieve our topology */
val = NULL;
OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_LOCAL_TOPO,
ORTE_PROC_MY_NAME, &val, OPAL_STRING);
if (OPAL_SUCCESS == ret && NULL != val) {
/* load the topology */
if (0 != hwloc_topology_init(&opal_hwloc_topology)) {
ret = OPAL_ERROR;
free(val);
error = "setting topology";
goto error;
}
if (0 != hwloc_topology_set_xmlbuffer(opal_hwloc_topology, val, strlen(val))) {
ret = OPAL_ERROR;
free(val);
hwloc_topology_destroy(opal_hwloc_topology);
error = "setting topology";
goto error;
}
/* since we are loading this from an external source, we have to
* explicitly set a flag so hwloc sets things up correctly
*/
if (0 != hwloc_topology_set_flags(opal_hwloc_topology,
(HWLOC_TOPOLOGY_FLAG_IS_THISSYSTEM |
HWLOC_TOPOLOGY_FLAG_WHOLE_SYSTEM |
HWLOC_TOPOLOGY_FLAG_IO_DEVICES))) {
ret = OPAL_ERROR;
hwloc_topology_destroy(opal_hwloc_topology);
free(val);
error = "setting topology";
goto error;
}
/* now load the topology */
if (0 != hwloc_topology_load(opal_hwloc_topology)) {
ret = OPAL_ERROR;
hwloc_topology_destroy(opal_hwloc_topology);
free(val);
error = "setting topology";
goto error;
}
free(val);
/* filter the cpus thru any default cpu set */
if (OPAL_SUCCESS != (ret = opal_hwloc_base_filter_cpus(opal_hwloc_topology))) {
error = "filtering topology";
goto error;
}
} else {
/* it wasn't passed down to us, so go get it */
if (OPAL_SUCCESS != (ret = opal_hwloc_base_get_topology())) {
error = "topology discovery";
goto error;
}
/* push it into the PMIx database in case someone
* tries to retrieve it so we avoid an attempt to
* get it again */
kv = OBJ_NEW(opal_value_t);
kv->key = strdup(OPAL_PMIX_LOCAL_TOPO);
kv->type = OPAL_STRING;
if (0 != (ret = hwloc_topology_export_xmlbuffer(opal_hwloc_topology, &kv->data.string, &u32))) {
error = "topology export";
goto error;
}
if (OPAL_SUCCESS != (ret = opal_pmix.store_local(ORTE_PROC_MY_NAME, kv))) {
error = "topology store";
goto error;
}
OBJ_RELEASE(kv);
}
/* get our local peers */
if (0 < orte_process_info.num_local_peers) {
/* if my local rank if too high, then that's an error */
if (orte_process_info.num_local_peers < orte_process_info.my_local_rank) {
ret = ORTE_ERR_BAD_PARAM;
error = "num local peers";
goto error;
}
/* retrieve the local peers */
OPAL_MODEX_RECV_VALUE(ret, OPAL_PMIX_LOCAL_PEERS,
ORTE_PROC_MY_NAME, &val, OPAL_STRING);
if (OPAL_SUCCESS == ret && NULL != val) {
peers = opal_argv_split(val, ',');
free(val);
/* and their cpusets, if available */
OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_LOCAL_CPUSETS, ORTE_PROC_MY_NAME, &val, OPAL_STRING);
if (OPAL_SUCCESS == ret && NULL != val) {
cpusets = opal_argv_split(val, ':');
free(val);
} else {
cpusets = NULL;
}
} else {
peers = NULL;
cpusets = NULL;
}
} else {
peers = NULL;
cpusets = NULL;
}
/* set the locality */
if (NULL != peers) {
/* indentify our cpuset */
if (NULL != cpusets) {
mycpuset = cpusets[orte_process_info.my_local_rank];
} else {
mycpuset = NULL;
}
name.jobid = ORTE_PROC_MY_NAME->jobid;
for (i=0; NULL != peers[i]; i++) {
kv = OBJ_NEW(opal_value_t);
kv->key = strdup(OPAL_PMIX_LOCALITY);
kv->type = OPAL_UINT16;
name.vpid = strtoul(peers[i], NULL, 10);
if (name.vpid == ORTE_PROC_MY_NAME->vpid) {
/* we are fully local to ourselves */
u16 = OPAL_PROC_ALL_LOCAL;
} else if (NULL == mycpuset || NULL == cpusets[i] ||
0 == strcmp(cpusets[i], "UNBOUND")) {
/* all we can say is that it shares our node */
u16 = OPAL_PROC_ON_CLUSTER | OPAL_PROC_ON_CU | OPAL_PROC_ON_NODE;
} else {
/* we have it, so compute the locality */
u16 = opal_hwloc_base_get_relative_locality(opal_hwloc_topology, mycpuset, cpusets[i]);
}
OPAL_OUTPUT_VERBOSE((1, orte_ess_base_framework.framework_output,
"%s ess:pmi:locality: proc %s locality %x",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&name), u16));
kv->data.uint16 = u16;
ret = opal_pmix.store_local(&name, kv);
if (OPAL_SUCCESS != ret) {
error = "local store of locality";
opal_argv_free(peers);
opal_argv_free(cpusets);
goto error;
}
OBJ_RELEASE(kv);
}
opal_argv_free(peers);
opal_argv_free(cpusets);
}
/* now that we have all required info, complete the setup */
if (ORTE_SUCCESS != (ret = orte_ess_base_app_setup(false))) {
ORTE_ERROR_LOG(ret);
error = "orte_ess_base_app_setup";
goto error;
}
/* setup process binding */
if (ORTE_SUCCESS != (ret = orte_ess_base_proc_binding())) {
error = "proc_binding";
goto error;
}
/* this needs to be set to enable debugger use when direct launched */
if (NULL == orte_process_info.my_daemon_uri) {
orte_standalone_operation = true;
}
/* set max procs */
if (orte_process_info.max_procs < orte_process_info.num_procs) {
orte_process_info.max_procs = orte_process_info.num_procs;
}
/*** PUSH DATA FOR OTHERS TO FIND ***/
/* push our RML URI in case others need to talk directly to us */
rmluri = orte_rml.get_contact_info();
/* push it out for others to use */
OPAL_MODEX_SEND_VALUE(ret, OPAL_PMIX_GLOBAL, OPAL_PMIX_PROC_URI, rmluri, OPAL_STRING);
if (ORTE_SUCCESS != ret) {
error = "pmix put uri";
goto error;
}
free(rmluri);
/* push our hostname so others can find us, if they need to */
OPAL_MODEX_SEND_VALUE(ret, OPAL_PMIX_GLOBAL, OPAL_PMIX_HOSTNAME, orte_process_info.nodename, OPAL_STRING);
if (ORTE_SUCCESS != ret) {
error = "db store hostname";
goto error;
}
/* if we are an ORTE app - and not an MPI app - then
* we need to exchange our connection info here.
* MPI_Init has its own modex, so we don't need to do
* two of them. However, if we don't do a modex at all,
* then processes have no way to communicate
*
* NOTE: only do this when the process originally launches.
* Cannot do this on a restart as the rest of the processes
* in the job won't be executing this step, so we would hang
*/
if (ORTE_PROC_IS_NON_MPI && !orte_do_not_barrier) {
opal_pmix.fence(NULL, 0);
}
return ORTE_SUCCESS;
error:
if (!progress_thread_running) {
/* can't send the help message, so ensure it
* comes out locally
*/
orte_show_help_finalize();
}
if (ORTE_ERR_SILENT != ret && !orte_report_silent_errors) {
orte_show_help("help-orte-runtime.txt",
"orte_init:startup:internal-failure",
true, error, ORTE_ERROR_NAME(ret), ret);
}
return ret;
}
static int allocate(orte_job_t *jdata, opal_list_t *nodes)
{
int i, n, val, dig, num_nodes;
orte_node_t *node;
#if OPAL_HAVE_HWLOC
orte_topology_t *t;
hwloc_topology_t topo;
hwloc_obj_t obj;
unsigned j, k;
struct hwloc_topology_support *support;
char **files=NULL;
char **topos = NULL;
bool use_local_topology = false;
#endif
char **node_cnt=NULL;
char **slot_cnt=NULL;
char **max_slot_cnt=NULL;
char *tmp;
char prefix[6];
node_cnt = opal_argv_split(mca_ras_simulator_component.num_nodes, ',');
if (NULL != mca_ras_simulator_component.slots) {
slot_cnt = opal_argv_split(mca_ras_simulator_component.slots, ',');
/* backfile the slot_cnt so every topology has a cnt */
tmp = slot_cnt[opal_argv_count(slot_cnt)-1];
for (n=opal_argv_count(slot_cnt); n < opal_argv_count(node_cnt); n++) {
opal_argv_append_nosize(&slot_cnt, tmp);
}
}
if (NULL != mca_ras_simulator_component.slots_max) {
max_slot_cnt = opal_argv_split(mca_ras_simulator_component.slots_max, ',');
/* backfill the max_slot_cnt as reqd */
tmp = max_slot_cnt[opal_argv_count(slot_cnt)-1];
for (n=opal_argv_count(max_slot_cnt); n < opal_argv_count(max_slot_cnt); n++) {
opal_argv_append_nosize(&max_slot_cnt, tmp);
}
}
#if OPAL_HAVE_HWLOC
if (NULL != mca_ras_simulator_component.topofiles) {
files = opal_argv_split(mca_ras_simulator_component.topofiles, ',');
if (opal_argv_count(files) != opal_argv_count(node_cnt)) {
orte_show_help("help-ras-base.txt", "ras-sim:mismatch", true);
goto error_silent;
}
} else if (NULL != mca_ras_simulator_component.topologies) {
topos = opal_argv_split(mca_ras_simulator_component.topologies, ',');
if (opal_argv_count(topos) != opal_argv_count(node_cnt)) {
orte_show_help("help-ras-base.txt", "ras-sim:mismatch", true);
goto error_silent;
}
} else {
/* use our topology */
use_local_topology = true;
}
#else
/* If we don't have hwloc and hwloc files were specified, then
error out (because we can't deliver that functionality) */
if (NULL == mca_ras_simulator_component.topofiles) {
orte_show_help("help-ras-simulator.txt",
"no hwloc support for topofiles", true);
goto error_silent;
}
#endif
/* setup the prefix to the node names */
snprintf(prefix, 6, "nodeA");
/* process the request */
for (n=0; NULL != node_cnt[n]; n++) {
num_nodes = strtol(node_cnt[n], NULL, 10);
/* get number of digits */
val = num_nodes;
for (dig=0; 0 != val; dig++) {
val /= 10;
}
/* set the prefix for this group of nodes */
prefix[4] += n;
/* check for topology */
#if OPAL_HAVE_HWLOC
if (use_local_topology) {
/* use our topology */
topo = opal_hwloc_topology;
} else if (NULL != files) {
if (0 != hwloc_topology_init(&topo)) {
orte_show_help("help-ras-simulator.txt",
"hwloc API fail", true,
__FILE__, __LINE__, "hwloc_topology_init");
goto error_silent;
}
if (0 != hwloc_topology_set_xml(topo, files[n])) {
orte_show_help("help-ras-simulator.txt",
"hwloc failed to load xml", true, files[n]);
hwloc_topology_destroy(topo);
goto error_silent;
}
/* since we are loading this from an external source, we have to
* explicitly set a flag so hwloc sets things up correctly
*/
if (0 != hwloc_topology_set_flags(topo, HWLOC_TOPOLOGY_FLAG_IS_THISSYSTEM)) {
orte_show_help("help-ras-simulator.txt",
"hwloc API fail", true,
__FILE__, __LINE__, "hwloc_topology_set_flags");
hwloc_topology_destroy(topo);
goto error_silent;
}
if (0 != hwloc_topology_load(topo)) {
orte_show_help("help-ras-simulator.txt",
"hwloc API fail", true,
__FILE__, __LINE__, "hwloc_topology_load");
hwloc_topology_destroy(topo);
goto error_silent;
}
/* remove the hostname from the topology. Unfortunately, hwloc
* decided to add the source hostname to the "topology", thus
* rendering it unusable as a pure topological description. So
* we remove that information here.
*/
obj = hwloc_get_root_obj(topo);
for (k=0; k < obj->infos_count; k++) {
if (NULL == obj->infos[k].name ||
NULL == obj->infos[k].value) {
continue;
}
if (0 == strncmp(obj->infos[k].name, "HostName", strlen("HostName"))) {
free(obj->infos[k].name);
free(obj->infos[k].value);
/* left justify the array */
for (j=k; j < obj->infos_count-1; j++) {
obj->infos[j] = obj->infos[j+1];
}
obj->infos[obj->infos_count-1].name = NULL;
obj->infos[obj->infos_count-1].value = NULL;
obj->infos_count--;
break;
}
}
/* unfortunately, hwloc does not include support info in its
* xml output :-(( To aid in debugging, we set it here
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(topo);
support->cpubind->set_thisproc_cpubind = mca_ras_simulator_component.have_cpubind;
support->membind->set_thisproc_membind = mca_ras_simulator_component.have_membind;
/* add it to our array */
t = OBJ_NEW(orte_topology_t);
t->topo = topo;
t->sig = opal_hwloc_base_get_topo_signature(topo);
opal_pointer_array_add(orte_node_topologies, t);
} else {
if (0 != hwloc_topology_init(&topo)) {
orte_show_help("help-ras-simulator.txt",
"hwloc API fail", true,
__FILE__, __LINE__, "hwloc_topology_init");
goto error_silent;
}
if (0 != hwloc_topology_set_synthetic(topo, topos[n])) {
orte_show_help("help-ras-simulator.txt",
"hwloc API fail", true,
__FILE__, __LINE__, "hwloc_topology_set_synthetic");
hwloc_topology_destroy(topo);
goto error_silent;
}
if (0 != hwloc_topology_load(topo)) {
orte_show_help("help-ras-simulator.txt",
"hwloc API fail", true,
__FILE__, __LINE__, "hwloc_topology_load");
hwloc_topology_destroy(topo);
goto error_silent;
}
if (OPAL_SUCCESS != opal_hwloc_base_filter_cpus(topo)) {
orte_show_help("help-ras-simulator.txt",
"hwloc API fail", true,
__FILE__, __LINE__, "opal_hwloc_base_filter_cpus");
hwloc_topology_destroy(topo);
goto error_silent;
}
/* remove the hostname from the topology. Unfortunately, hwloc
* decided to add the source hostname to the "topology", thus
* rendering it unusable as a pure topological description. So
* we remove that information here.
*/
obj = hwloc_get_root_obj(topo);
for (k=0; k < obj->infos_count; k++) {
if (NULL == obj->infos[k].name ||
NULL == obj->infos[k].value) {
continue;
}
if (0 == strncmp(obj->infos[k].name, "HostName", strlen("HostName"))) {
free(obj->infos[k].name);
free(obj->infos[k].value);
/* left justify the array */
for (j=k; j < obj->infos_count-1; j++) {
obj->infos[j] = obj->infos[j+1];
}
obj->infos[obj->infos_count-1].name = NULL;
obj->infos[obj->infos_count-1].value = NULL;
obj->infos_count--;
break;
}
}
/* unfortunately, hwloc does not include support info in its
* xml output :-(( To aid in debugging, we set it here
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(topo);
support->cpubind->set_thisproc_cpubind = mca_ras_simulator_component.have_cpubind;
support->membind->set_thisproc_membind = mca_ras_simulator_component.have_membind;
/* add it to our array */
t = OBJ_NEW(orte_topology_t);
t->topo = topo;
t->sig = opal_hwloc_base_get_topo_signature(topo);
opal_pointer_array_add(orte_node_topologies, t);
}
#endif
for (i=0; i < num_nodes; i++) {
node = OBJ_NEW(orte_node_t);
asprintf(&node->name, "%s%0*d", prefix, dig, i);
node->state = ORTE_NODE_STATE_UP;
node->slots_inuse = 0;
#if OPAL_HAVE_HWLOC
if (NULL == max_slot_cnt || NULL == max_slot_cnt[n]) {
node->slots_max = 0;
} else {
obj = hwloc_get_root_obj(topo);
node->slots_max = opal_hwloc_base_get_npus(topo, obj);
}
if (NULL == slot_cnt || NULL == slot_cnt[n]) {
node->slots = 0;
} else {
obj = hwloc_get_root_obj(topo);
node->slots = opal_hwloc_base_get_npus(topo, obj);
}
node->topology = topo;
#endif
opal_output_verbose(1, orte_ras_base_framework.framework_output,
"Created Node <%10s> [%3d : %3d]",
node->name, node->slots, node->slots_max);
opal_list_append(nodes, &node->super);
}
}
/* record the number of allocated nodes */
orte_num_allocated_nodes = opal_list_get_size(nodes);
if (NULL != max_slot_cnt) {
opal_argv_free(max_slot_cnt);
}
if (NULL != slot_cnt) {
opal_argv_free(slot_cnt);
}
if (NULL != node_cnt) {
opal_argv_free(node_cnt);
}
return ORTE_SUCCESS;
error_silent:
if (NULL != max_slot_cnt) {
opal_argv_free(max_slot_cnt);
}
if (NULL != slot_cnt) {
opal_argv_free(slot_cnt);
}
if (NULL != node_cnt) {
opal_argv_free(node_cnt);
}
return ORTE_ERR_SILENT;
}
int main(int argc, char* argv[])
{
hwloc_obj_t obj;
unsigned j, k;
struct hwloc_topology_support *support;
int rc;
if (2 != argc) {
fprintf(stderr, "Usage: opal_hwloc <topofile>\n");
exit(1);
}
if (0 > (rc = opal_init(&argc, &argv))) {
fprintf(stderr, "opal_hwloc: couldn't init opal - error code %d\n", rc);
return rc;
}
if (0 != hwloc_topology_init(&my_topology)) {
return OPAL_ERR_NOT_SUPPORTED;
}
if (0 != hwloc_topology_set_xml(my_topology, argv[1])) {
hwloc_topology_destroy(my_topology);
return OPAL_ERR_NOT_SUPPORTED;
}
/* since we are loading this from an external source, we have to
* explicitly set a flag so hwloc sets things up correctly
*/
if (0 != hwloc_topology_set_flags(my_topology,
(HWLOC_TOPOLOGY_FLAG_IS_THISSYSTEM |
HWLOC_TOPOLOGY_FLAG_WHOLE_SYSTEM |
HWLOC_TOPOLOGY_FLAG_IO_DEVICES))) {
hwloc_topology_destroy(my_topology);
return OPAL_ERR_NOT_SUPPORTED;
}
if (0 != hwloc_topology_load(my_topology)) {
hwloc_topology_destroy(my_topology);
return OPAL_ERR_NOT_SUPPORTED;
}
/* remove the hostname from the topology. Unfortunately, hwloc
* decided to add the source hostname to the "topology", thus
* rendering it unusable as a pure topological description. So
* we remove that information here.
*/
obj = hwloc_get_root_obj(my_topology);
for (k=0; k < obj->infos_count; k++) {
if (NULL == obj->infos[k].name ||
NULL == obj->infos[k].value) {
continue;
}
if (0 == strncmp(obj->infos[k].name, "HostName", strlen("HostName"))) {
free(obj->infos[k].name);
free(obj->infos[k].value);
/* left justify the array */
for (j=k; j < obj->infos_count-1; j++) {
obj->infos[j] = obj->infos[j+1];
}
obj->infos[obj->infos_count-1].name = NULL;
obj->infos[obj->infos_count-1].value = NULL;
obj->infos_count--;
break;
}
}
/* unfortunately, hwloc does not include support info in its
* xml output :-(( We default to assuming it is present as
* systems that use this option are likely to provide
* binding support
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(my_topology);
support->cpubind->set_thisproc_cpubind = true;
/* filter the cpus thru any default cpu set */
opal_hwloc_base_filter_cpus(my_topology);
/* fill opal_cache_line_size global with the smallest L1 cache
line size */
fill_cache_line_size();
/* test it */
if (NULL == hwloc_get_obj_by_type(my_topology, HWLOC_OBJ_CORE, 0)) {
fprintf(stderr, "DIDN'T FIND A CORE\n");
}
hwloc_topology_destroy(my_topology);
opal_finalize();
return 0;
}
