If the RTE fails to deliver the daemon information,

gracefully fallback to a non-reordered communicator.
Optimize the loops building the process hierarchy.

ompi/mca/topo/treematch/topo_treematch_dist_graph_create.c

@@ -44,9 +44,9 @@
     while(0);
 
 #define FALLBACK()                  \
-    do { free(nodes_roots);			\
+    do { free(nodes_roots);         \
         free(local_procs);          \
-        hwloc_bitmap_free(set);     \
+        if( NULL != set) hwloc_bitmap_free(set);     \
         goto fallback; }            \
     while(0);
 
@@ -181,19 +181,16 @@ int mca_topo_treematch_dist_graph_create(mca_topo_base_module_t* topo_module,
             num_procs_in_node++;
     }
 
-    /* Get the ranks of the local procs in comm_old */
+    vpids = (int *)malloc(size * sizeof(int));
+    colors = (int *)malloc(size * sizeof(int));
     local_procs = (int *)malloc(num_procs_in_node * sizeof(int));
     for(i = idx = 0 ; i < size ; i++){
         proc = ompi_group_peer_lookup(comm_old->c_local_group, i);
         if (( i == rank ) ||
-            (OPAL_PROC_ON_LOCAL_NODE(proc->super.proc_flags)))
+            (OPAL_PROC_ON_LOCAL_NODE(proc->super.proc_flags))) {
             local_procs[idx++] = i;
-    }
+        }
 
-    vpids = (int *)malloc(size * sizeof(int));
-    colors = (int *)malloc(size * sizeof(int));
-    for(i = 0; i < size ; i++) {
-        proc = ompi_group_peer_lookup(comm_old->c_local_group, i);
         pval = &val;
         OPAL_MODEX_RECV_VALUE(err, OPAL_PMIX_NODEID, &(proc->super.proc_name), &pval, OPAL_UINT32);
         if( OPAL_SUCCESS != err ) {
@@ -220,22 +217,30 @@ int mca_topo_treematch_dist_graph_create(mca_topo_base_module_t* topo_module,
     }
 #endif
     /* clean-up dupes in the array */
-    for(i = 0; i < size ; i++)
+    for(i = 0; i < size; i++) {
-        if ( -1 ==  vpids[i] )
+        if( -1 ==  vpids[i] )
             continue;
-        else
+
-            for(j = i+1 ; j < size ; j++)
+        num_nodes++; /* update the number of nodes */
-                if( vpids[j] != -1 )
+
-                    if( vpids[i] == vpids[j] )
+        for(j = i+1; j < size; j++)
-                        vpids[j] = -1;
+            if( vpids[j] != -1 )
-    /* compute number of nodes */
+                if( vpids[i] == vpids[j] )
-    for(i = 0; i < size ; i++)
+                    vpids[j] = -1;
-        if( vpids[i] != -1 )
+    }
-            num_nodes++;
+    if( 0 == num_nodes ) {
+        /* No useful info has been retrieved from the runtime. Fallback
+         * and create a duplicate of the original communicator */
+        free(vpids);
+        free(colors);
+        free(local_procs);
+        err = OMPI_SUCCESS;  /* return with success */
+        goto fallback;
+    }
     /* compute local roots ranks in comm_old */
     /* Only the global root needs to do this */
     if(0 == rank) {
-        nodes_roots = (int *)calloc(num_nodes,sizeof(int));
+        nodes_roots = (int *)calloc(num_nodes, sizeof(int));
         for(i = idx = 0; i < size ; i++)
             if( vpids[i] != -1 )
                 nodes_roots[idx++] = i;