static int rte_init(void) { int rc, ret; char *error = NULL; char *envar, *ev1, *ev2; uint64_t unique_key[2]; char *string_key; opal_value_t *kv; char *val; int u32, *u32ptr; uint16_t u16, *u16ptr; orte_process_name_t name; /* run the prolog */ if (ORTE_SUCCESS != (rc = orte_ess_base_std_prolog())) { ORTE_ERROR_LOG(rc); return rc; } u32ptr = &u32; u16ptr = &u16; if (NULL != mca_ess_singleton_component.server_uri) { /* we are going to connect to a server HNP */ if (0 == strncmp(mca_ess_singleton_component.server_uri, "file", strlen("file")) || 0 == strncmp(mca_ess_singleton_component.server_uri, "FILE", strlen("FILE"))) { char input[1024], *filename; FILE *fp; /* it is a file - get the filename */ filename = strchr(mca_ess_singleton_component.server_uri, ':'); if (NULL == filename) { /* filename is not correctly formatted */ orte_show_help("help-orterun.txt", "orterun:ompi-server-filename-bad", true, "singleton", mca_ess_singleton_component.server_uri); return ORTE_ERROR; } ++filename; /* space past the : */ if (0 >= strlen(filename)) { /* they forgot to give us the name! */ orte_show_help("help-orterun.txt", "orterun:ompi-server-filename-missing", true, "singleton", mca_ess_singleton_component.server_uri); return ORTE_ERROR; } /* open the file and extract the uri */ fp = fopen(filename, "r"); if (NULL == fp) { /* can't find or read file! */ orte_show_help("help-orterun.txt", "orterun:ompi-server-filename-access", true, "singleton", mca_ess_singleton_component.server_uri); return ORTE_ERROR; } memset(input, 0, 1024); // initialize the array to ensure a NULL termination if (NULL == fgets(input, 1023, fp)) { /* something malformed about file */ fclose(fp); orte_show_help("help-orterun.txt", "orterun:ompi-server-file-bad", true, "singleton", mca_ess_singleton_component.server_uri, "singleton"); return ORTE_ERROR; } fclose(fp); input[strlen(input)-1] = '\0'; /* remove newline */ orte_process_info.my_hnp_uri = strdup(input); } else { orte_process_info.my_hnp_uri = strdup(mca_ess_singleton_component.server_uri); } /* save the daemon uri - we will process it later */ orte_process_info.my_daemon_uri = strdup(orte_process_info.my_hnp_uri); /* construct our name - we are in their job family, so we know that * much. However, we cannot know how many other singletons and jobs * this HNP is running. Oh well - if someone really wants to use this * option, they can try to figure it out. For now, we'll just assume * we are the only ones */ ORTE_PROC_MY_NAME->jobid = ORTE_CONSTRUCT_LOCAL_JOBID(ORTE_PROC_MY_HNP->jobid, 1); /* obviously, we are vpid=0 for this job */ ORTE_PROC_MY_NAME->vpid = 0; /* for convenience, push the pubsub version of this param into the environ */ opal_setenv (OPAL_MCA_PREFIX"pubsub_orte_server", orte_process_info.my_hnp_uri, true, &environ); } else if (NULL != getenv("SINGULARITY_CONTAINER") || mca_ess_singleton_component.isolated) { /* ensure we use the isolated pmix component */ opal_setenv (OPAL_MCA_PREFIX"pmix", "isolated", true, &environ); } else { /* spawn our very own HNP to support us */ if (ORTE_SUCCESS != (rc = fork_hnp())) { ORTE_ERROR_LOG(rc); return rc; } /* our name was given to us by the HNP */ opal_setenv (OPAL_MCA_PREFIX"pmix", "^s1,s2,cray,isolated", true, &environ); } /* 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))) { error = "opening pmix"; goto error; } if (OPAL_SUCCESS != (ret = opal_pmix_base_select())) { error = "select pmix"; 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; } /* 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; name.jobid = OPAL_PROC_MY_NAME.jobid; name.vpid = ORTE_VPID_WILDCARD; /* 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, &name, &u32ptr, OPAL_UINT32); if (OPAL_SUCCESS != ret) { error = "getting max procs"; goto error; } orte_process_info.max_procs = u32; /* we are a singleton, so there is only one proc in the job */ orte_process_info.num_procs = 1; /* 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(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; } /* set some other standard values */ 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; } 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 */ OPAL_MODEX_RECV_VALUE(ret, OPAL_PMIX_LOCAL_TOPO, &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); } 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); } /* use the std app init to complete the procedure */ if (ORTE_SUCCESS != (rc = orte_ess_base_app_setup(true))) { ORTE_ERROR_LOG(rc); return rc; } /* 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; } return ORTE_SUCCESS; error: 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; }
int orte_ess_base_proc_binding(void) { hwloc_obj_t node, obj; hwloc_cpuset_t cpus, nodeset; hwloc_obj_type_t target; unsigned int cache_level = 0; struct hwloc_topology_support *support; char *map; int ret; char *error=NULL; hwloc_cpuset_t mycpus; /* Determine if we were pre-bound or not */ if (NULL != getenv(OPAL_MCA_PREFIX"orte_bound_at_launch")) { orte_proc_is_bound = true; if (NULL != (map = getenv(OPAL_MCA_PREFIX"orte_base_applied_binding"))) { orte_proc_applied_binding = hwloc_bitmap_alloc(); if (0 != (ret = hwloc_bitmap_list_sscanf(orte_proc_applied_binding, map))) { error = "applied_binding parse"; goto error; } } } /* see if we were bound when launched */ if (!orte_proc_is_bound) { OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output, "%s Not bound at launch", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME))); /* we were not bound at launch */ if (NULL == opal_hwloc_topology) { /* there is nothing we can do, so just return */ return ORTE_SUCCESS; } support = (struct hwloc_topology_support*)hwloc_topology_get_support(opal_hwloc_topology); /* get our node object */ node = hwloc_get_root_obj(opal_hwloc_topology); nodeset = opal_hwloc_base_get_available_cpus(opal_hwloc_topology, node); /* get our bindings */ cpus = hwloc_bitmap_alloc(); if (hwloc_get_cpubind(opal_hwloc_topology, cpus, HWLOC_CPUBIND_PROCESS) < 0) { /* we are NOT bound if get_cpubind fails, nor can we be bound - the * environment does not support it */ hwloc_bitmap_free(cpus); OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output, "%s Binding not supported", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME))); goto MOVEON; } /* we are bound if the two cpusets are not equal, * or if there is only ONE cpu available to us */ if (0 != hwloc_bitmap_compare(cpus, nodeset) || opal_hwloc_base_single_cpu(nodeset) || opal_hwloc_base_single_cpu(cpus)) { /* someone external set it - indicate it is set * so that we know */ orte_proc_is_bound = true; hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, cpus); hwloc_bitmap_free(cpus); OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output, "%s Process was externally bound", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME))); } else if (support->cpubind->set_thisproc_cpubind && OPAL_BINDING_POLICY_IS_SET(opal_hwloc_binding_policy) && OPAL_BIND_TO_NONE != OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) { /* the system is capable of doing processor affinity, but it * has not yet been set - see if a slot_list was given */ hwloc_bitmap_zero(cpus); if (OPAL_BIND_TO_CPUSET == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) { if (OPAL_SUCCESS != (ret = opal_hwloc_base_slot_list_parse(opal_hwloc_base_slot_list, opal_hwloc_topology, OPAL_HWLOC_LOGICAL, cpus))) { error = "Setting processor affinity failed"; hwloc_bitmap_free(cpus); goto error; } if (0 > hwloc_set_cpubind(opal_hwloc_topology, cpus, 0)) { error = "Setting processor affinity failed"; hwloc_bitmap_free(cpus); goto error; } hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, cpus); hwloc_bitmap_free(cpus); orte_proc_is_bound = true; OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output, "%s Process bound according to slot_list", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME))); } else { /* cleanup */ hwloc_bitmap_free(cpus); /* get the node rank */ if (ORTE_NODE_RANK_INVALID == orte_process_info.my_node_rank) { /* this is not an error - could be due to being * direct launched - so just ignore and leave * us unbound */ OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output, "%s Process not bound - no node rank available", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME))); goto MOVEON; } /* if the binding policy is hwthread, then we bind to the nrank-th * hwthread on this node */ if (OPAL_BIND_TO_HWTHREAD == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) { if (NULL == (obj = opal_hwloc_base_get_obj_by_type(opal_hwloc_topology, HWLOC_OBJ_PU, 0, orte_process_info.my_node_rank, OPAL_HWLOC_LOGICAL))) { ret = ORTE_ERR_NOT_FOUND; error = "Getting hwthread object"; goto error; } cpus = opal_hwloc_base_get_available_cpus(opal_hwloc_topology, obj); if (0 > hwloc_set_cpubind(opal_hwloc_topology, cpus, 0)) { ret = ORTE_ERROR; error = "Setting processor affinity failed"; goto error; } hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, cpus); hwloc_bitmap_free(cpus); OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output, "%s Process bound to hwthread", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME))); } else if (OPAL_BIND_TO_CORE == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) { /* if the binding policy is core, then we bind to the nrank-th * core on this node */ if (NULL == (obj = opal_hwloc_base_get_obj_by_type(opal_hwloc_topology, HWLOC_OBJ_CORE, 0, orte_process_info.my_node_rank, OPAL_HWLOC_LOGICAL))) { ret = ORTE_ERR_NOT_FOUND; error = "Getting core object"; goto error; } cpus = opal_hwloc_base_get_available_cpus(opal_hwloc_topology, obj); if (0 > hwloc_set_cpubind(opal_hwloc_topology, cpus, 0)) { error = "Setting processor affinity failed"; ret = ORTE_ERROR; goto error; } hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, cpus); OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output, "%s Process bound to core", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME))); } else { /* for all higher binding policies, we bind to the specified * object that the nrank-th core belongs to */ if (NULL == (obj = opal_hwloc_base_get_obj_by_type(opal_hwloc_topology, HWLOC_OBJ_CORE, 0, orte_process_info.my_node_rank, OPAL_HWLOC_LOGICAL))) { ret = ORTE_ERR_NOT_FOUND; error = "Getting core object"; goto error; } if (OPAL_BIND_TO_L1CACHE == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) { target = HWLOC_OBJ_CACHE; cache_level = 1; } else if (OPAL_BIND_TO_L2CACHE == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) { target = HWLOC_OBJ_CACHE; cache_level = 2; } else if (OPAL_BIND_TO_L3CACHE == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) { target = HWLOC_OBJ_CACHE; cache_level = 3; } else if (OPAL_BIND_TO_SOCKET == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) { target = HWLOC_OBJ_SOCKET; } else if (OPAL_BIND_TO_NUMA == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy)) { target = HWLOC_OBJ_NODE; } else { ret = ORTE_ERR_NOT_FOUND; error = "Binding policy not known"; goto error; } for (obj = obj->parent; NULL != obj; obj = obj->parent) { if (target == obj->type) { if (HWLOC_OBJ_CACHE == target && cache_level != obj->attr->cache.depth) { continue; } /* this is the place! */ cpus = opal_hwloc_base_get_available_cpus(opal_hwloc_topology, obj); if (0 > hwloc_set_cpubind(opal_hwloc_topology, cpus, 0)) { ret = ORTE_ERROR; error = "Setting processor affinity failed"; goto error; } hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, cpus); orte_proc_is_bound = true; OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output, "%s Process bound to %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), hwloc_obj_type_string(target))); break; } } if (!orte_proc_is_bound) { ret = ORTE_ERROR; error = "Setting processor affinity failed"; goto error; } } } } } else { OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output, "%s Process bound at launch", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME))); } MOVEON: /* get or update our local cpuset - it will get used multiple * times, so it's more efficient to keep a global copy */ opal_hwloc_base_get_local_cpuset(); /* get the cpus we are bound to */ mycpus = hwloc_bitmap_alloc(); if (hwloc_get_cpubind(opal_hwloc_topology, mycpus, HWLOC_CPUBIND_PROCESS) < 0) { if (NULL != orte_process_info.cpuset) { free(orte_process_info.cpuset); orte_process_info.cpuset = NULL; } if (opal_hwloc_report_bindings || 4 < opal_output_get_verbosity(orte_ess_base_framework.framework_output)) { opal_output(0, "MCW rank %d is not bound", ORTE_PROC_MY_NAME->vpid); } } else { /* store/update the string representation of our local binding */ if (NULL != orte_process_info.cpuset) { free(orte_process_info.cpuset); orte_process_info.cpuset = NULL; } hwloc_bitmap_list_asprintf(&orte_process_info.cpuset, mycpus); /* report the binding, if requested */ if (opal_hwloc_report_bindings || 4 < opal_output_get_verbosity(orte_ess_base_framework.framework_output)) { char tmp1[1024], tmp2[1024]; if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2str(tmp1, sizeof(tmp1), opal_hwloc_topology, mycpus)) { opal_output(0, "MCW rank %d is not bound (or bound to all available processors)", ORTE_PROC_MY_NAME->vpid); } else { opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), opal_hwloc_topology, mycpus); opal_output(0, "MCW rank %d bound to %s: %s", ORTE_PROC_MY_NAME->vpid, tmp1, tmp2); } } } hwloc_bitmap_free(mycpus); /* push our cpuset so others can calculate our locality */ if (NULL != orte_process_info.cpuset) { OPAL_MODEX_SEND_VALUE(ret, OPAL_PMIX_GLOBAL, OPAL_PMIX_CPUSET, orte_process_info.cpuset, OPAL_STRING); } return ORTE_SUCCESS; error: if (ORTE_ERR_SILENT != ret) { orte_show_help("help-orte-runtime", "orte_init:startup:internal-failure", true, error, ORTE_ERROR_NAME(ret), ret); } return ORTE_ERR_SILENT; }
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 rte_init(void) { int ret; char *error = NULL; char *envar, *ev1, *ev2; uint64_t unique_key[2]; char *string_key; opal_value_t *kv; char *val; int u32, *u32ptr; uint16_t u16, *u16ptr; char **peers=NULL, *mycpuset; opal_process_name_t wildcard_rank, pname; bool bool_val, *bool_ptr = &bool_val, tdir_mca_override = false; 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(NULL)))) { /* we cannot run - this could be due to being direct launched * without the required PMI support being built. Try to detect * that scenario and warn the user */ if (ORTE_SCHIZO_DIRECT_LAUNCHED == orte_schizo.check_launch_environment() && NULL != (envar = getenv("ORTE_SCHIZO_DETECTION"))) { if (0 == strcmp(envar, "SLURM")) { /* yes to both - so emit a hopefully helpful * error message and abort */ orte_show_help_finalize(); orte_show_help("help-ess-base.txt", "slurm-error", true); return ORTE_ERR_SILENT; } else if (0 == strcmp(envar, "ALPS")) { /* we were direct launched by ALPS */ orte_show_help_finalize(); orte_show_help("help-ess-base.txt", "alps-error", true); return ORTE_ERR_SILENT; } } 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; /* setup a name for retrieving data associated with the job */ wildcard_rank.jobid = ORTE_PROC_MY_NAME->jobid; wildcard_rank.vpid = ORTE_NAME_WILDCARD->vpid; /* setup a name for retrieving proc-specific data */ pname.jobid = ORTE_PROC_MY_NAME->jobid; pname.vpid = 0; /* 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 for this application */ OPAL_MODEX_RECV_VALUE(ret, OPAL_PMIX_MAX_PROCS, &wildcard_rank, &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, &wildcard_rank, &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, &wildcard_rank, &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; } /* get number of nodes in the job */ OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_NUM_NODES, &wildcard_rank, &u32ptr, OPAL_UINT32); if (OPAL_SUCCESS == ret) { orte_process_info.num_nodes = u32; } /* 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 temp directories info */ OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_TMPDIR, &wildcard_rank, &val, OPAL_STRING); if (OPAL_SUCCESS == ret && NULL != val) { /* We want to provide user with ability * to override RM settings at his own risk */ if( NULL == orte_process_info.top_session_dir ){ orte_process_info.top_session_dir = val; } else { /* keep the MCA setting */ tdir_mca_override = true; free(val); } val = NULL; } if( !tdir_mca_override ){ OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_NSDIR, &wildcard_rank, &val, OPAL_STRING); if (OPAL_SUCCESS == ret && NULL != val) { /* We want to provide user with ability * to override RM settings at his own risk */ if( NULL == orte_process_info.job_session_dir ){ orte_process_info.job_session_dir = val; } else { /* keep the MCA setting */ free(val); tdir_mca_override = true; } val = NULL; } } if( !tdir_mca_override ){ OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_PROCDIR, &wildcard_rank, &val, OPAL_STRING); if (OPAL_SUCCESS == ret && NULL != val) { /* We want to provide user with ability * to override RM settings at his own risk */ if( NULL == orte_process_info.proc_session_dir ){ orte_process_info.proc_session_dir = val; } else { /* keep the MCA setting */ tdir_mca_override = true; free(val); } val = NULL; } } if( !tdir_mca_override ){ OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_TDIR_RMCLEAN, &wildcard_rank, &bool_ptr, OPAL_BOOL); if (OPAL_SUCCESS == ret ) { orte_process_info.rm_session_dirs = bool_val; } } /* 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, &wildcard_rank, &val, OPAL_STRING); if (OPAL_SUCCESS == ret && NULL != val) { peers = opal_argv_split(val, ','); free(val); } else { peers = NULL; } } else { peers = NULL; } /* set the locality */ if (NULL != peers) { /* identify our location */ val = NULL; OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_LOCALITY_STRING, ORTE_PROC_MY_NAME, &val, OPAL_STRING); if (OPAL_SUCCESS == ret && NULL != val) { mycpuset = val; } else { mycpuset = NULL; } pname.jobid = ORTE_PROC_MY_NAME->jobid; for (i=0; NULL != peers[i]; i++) { pname.vpid = strtoul(peers[i], NULL, 10); if (pname.vpid == ORTE_PROC_MY_NAME->vpid) { /* we are fully local to ourselves */ u16 = OPAL_PROC_ALL_LOCAL; } else { val = NULL; OPAL_MODEX_RECV_VALUE_OPTIONAL(ret, OPAL_PMIX_LOCALITY_STRING, &pname, &val, OPAL_STRING); if (OPAL_SUCCESS == ret && NULL != val) { u16 = opal_hwloc_compute_relative_locality(mycpuset, val); free(val); } else { /* all we can say is that it shares our node */ u16 = OPAL_PROC_ON_CLUSTER | OPAL_PROC_ON_CU | OPAL_PROC_ON_NODE; } } kv = OBJ_NEW(opal_value_t); kv->key = strdup(OPAL_PMIX_LOCALITY); kv->type = OPAL_UINT16; OPAL_OUTPUT_VERBOSE((1, orte_ess_base_framework.framework_output, "%s ess:pmi:locality: proc %s locality %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_NAME_PRINT(&pname), opal_hwloc_base_print_locality(u16))); kv->data.uint16 = u16; ret = opal_pmix.store_local(&pname, kv); if (OPAL_SUCCESS != ret) { error = "local store of locality"; opal_argv_free(peers); if (NULL != mycpuset) { free(mycpuset); } goto error; } OBJ_RELEASE(kv); } opal_argv_free(peers); if (NULL != mycpuset) { free(mycpuset); } } /* 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 our hostname so others can find us, if they need to - the * native PMIx component will ignore this request as the hostname * is provided by the system */ 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) { /* need to commit the data before we fence */ opal_pmix.commit(); 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; }