utils/calc: fix searching CPU objects below heterogeneous NUMA

socket:0.numa:0.core:0 looks for first core whose cpuset and nodeset
are included in socket:0.numa:0. This fails on heterogeneous memory
machines since this core may have 2 local NUMAs in its nodeset,
hence not included in socket:0.numa:0's nodeset.

Fix this by checking if the nodeset intersects instead of being included.

Note that the empty bitmap is isincluded in any bitmap, while it doesn't
intersect with anything, hence we have to check empty nodesets too.

Thanks to Antoine Morvan for the report.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

Author: bgoglin

utils/hwloc/hwloc-calc.h

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2022 Inria.  All rights reserved.
+ * Copyright © 2009-2023 Inria.  All rights reserved.
  * Copyright © 2009-2012 Université Bordeaux
  * Copyright © 2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -88,7 +88,7 @@ hwloc_calc_get_nbobjs_inside_sets_by_depth(struct hwloc_calc_location_context_s
   while ((obj = hwloc_get_next_obj_by_depth(topology, depth, obj)) != NULL) {
     if (!hwloc_bitmap_isincluded(obj->cpuset, cpuset))
       continue;
-    if (!hwloc_bitmap_isincluded(obj->nodeset, nodeset))
+    if (!hwloc_bitmap_iszero(obj->nodeset) && !hwloc_bitmap_intersects(obj->nodeset, nodeset))
       continue;
     if (hwloc_bitmap_iszero(obj->cpuset) && hwloc_bitmap_iszero(obj->nodeset))
       /* ignore objects with empty sets (both can be empty when outside of cgroup) */
@@ -117,7 +117,7 @@ hwloc_calc_get_obj_inside_sets_by_depth(struct hwloc_calc_location_context_s *lc
   while ((obj = hwloc_get_next_obj_by_depth(topology, depth, obj)) != NULL) {
     if (!hwloc_bitmap_isincluded(obj->cpuset, cpuset))
       continue;
-    if (!hwloc_bitmap_isincluded(obj->nodeset, nodeset))
+    if (!hwloc_bitmap_iszero(obj->nodeset) && !hwloc_bitmap_intersects(obj->nodeset, nodeset))
       continue;
     if (hwloc_bitmap_iszero(obj->cpuset) && hwloc_bitmap_iszero(obj->nodeset))
       /* ignore objects with empty sets (both can be empty when outside of cgroup) */

utils/hwloc/test-hwloc-calc.output

@@ -130,6 +130,9 @@ PU:22_PU:23_Core:6_Core:7_Group0:2
 # Converting NUMA Nodes from logical to physical
 0,2
 
+# Search cores within single NUMA when they have multiple local NUMA nodes
+0x00000060
+
 # Physical output of NUMA Nodes when out-of-order in the topology
 2,1
 

utils/hwloc/test-hwloc-calc.sh.in

@@ -181,6 +181,10 @@ set -e
   $calc --if synthetic --input "node:4 core:4 pu:4" --ni 0x5 -I node --po
   echo
 
+  echo "# Search cores within single NUMA when they have multiple local NUMA nodes"
+  $calc --if synthetic --input "pack:2 [numa] [numa] core:4 pu:1" sock:1.numa:1.core:1-2
+  echo
+
   echo "# Physical output of NUMA Nodes when out-of-order in the topology"
   $calc --if synthetic --input "node:4(indexes=3,2,1,0) pu:2" node:1-2 --po -I node
   echo