Merge pull request #1250 from rhc54/topic/rf

Fix the default slot mapping in rank file mapper

orte/mca/rmaps/rank_file/rmaps_rank_file.c

@@ -196,10 +196,42 @@ static int orte_rmaps_rf_map(orte_job_t *jdata)
             rank = vpid_start + k;
             /* get the rankfile entry for this rank */
             if (NULL == (rfmap = (orte_rmaps_rank_file_map_t*)opal_pointer_array_get_item(&rankmap, rank))) {
+                /* if we were give a default slot-list, then use it */
+                if (NULL != opal_hwloc_base_slot_list) {
+                    slots = opal_hwloc_base_slot_list;
+                    /* take the next node off of the available list */
+                    node = NULL;
+                    OPAL_LIST_FOREACH(nd, &node_list, orte_node_t) {
+                        /* if adding one to this node would oversubscribe it, then try
+                         * the next one */
+                        if (nd->slots <= (int)nd->num_procs) {
+                            continue;
+                        }
+                        /* take this one */
+                        node = nd;
+                        break;
+                    }
+                    if (NULL == node) {
+                        /* all would be oversubscribed, so take the least loaded one */
+                        k = UINT32_MAX;
+                        OPAL_LIST_FOREACH(nd, &node_list, orte_node_t) {
+                            if (nd->num_procs < k) {
+                                k = nd->num_procs;
+                                node = nd;
+                            }
+                        }
+                    }
+                    /* if we still have nothing, then something is very wrong */
+                    if (NULL == node) {
+                        rc = ORTE_ERR_OUT_OF_RESOURCE;
+                        goto error;
+                    }
+                } else {
                     /* all ranks must be specified */
                     orte_show_help("help-rmaps_rank_file.txt", "missing-rank", true, rank, orte_rankfile);
                     rc = ORTE_ERR_SILENT;
                     goto error;
+                }
             } else {
                 if (0 == strlen(rfmap->slot_list)) {
                     /* rank was specified but no slot list given - that's an error */
@@ -208,8 +240,6 @@ static int orte_rmaps_rf_map(orte_job_t *jdata)
                     goto error;
                 }
                 slots = rfmap->slot_list;
-            }
-
                 /* find the node where this proc was assigned */
                 node = NULL;
                 OPAL_LIST_FOREACH(nd, &node_list, orte_node_t) {
@@ -235,7 +265,7 @@ static int orte_rmaps_rf_map(orte_job_t *jdata)
                         node = root_node;
                         break;
                     }
-
+                }
             }
             if (NULL == node) {
                 orte_show_help("help-rmaps_rank_file.txt","bad-host", true, rfmap->node_name);