distances: don't enforce the matrix name for NVLink operations

It may work on custom matrices
as long as ports are subtype'd "NVSwitch" for now.

Makes #705 obsolete

Signed-off-by: Brice Goglin <[email protected]>

Author: bgoglin

hwloc/distances.c

@@ -1386,19 +1386,12 @@ static __hwloc_inline int is_nvswitch(hwloc_obj_t obj)
 }
 
 static int
-hwloc__distances_transform_merge_switch_ports(hwloc_topology_t topology,
-                                              struct hwloc_distances_s *distances)
+hwloc__distances_transform_merge_switch_ports(struct hwloc_distances_s *distances)
 {
-  struct hwloc_internal_distances_s *dist = hwloc__internal_distances_from_public(topology, distances);
   hwloc_obj_t *objs = distances->objs;
   hwloc_uint64_t *values = distances->values;
   unsigned first, i, j, nbobjs = distances->nbobjs;
 
-  if (!dist || strcmp(dist->name, "NVLinkBandwidth")) {
-    errno = EINVAL;
-    return -1;
-  }
-
   /* find the first port */
   first = (unsigned) -1;
   for(i=0; i<nbobjs; i++)
@@ -1434,20 +1427,13 @@ hwloc__distances_transform_merge_switch_ports(hwloc_topology_t topology,
 }
 
 static int
-hwloc__distances_transform_transitive_closure(hwloc_topology_t topology,
-                                              struct hwloc_distances_s *distances)
+hwloc__distances_transform_transitive_closure(struct hwloc_distances_s *distances)
 {
-  struct hwloc_internal_distances_s *dist = hwloc__internal_distances_from_public(topology, distances);
   hwloc_obj_t *objs = distances->objs;
   hwloc_uint64_t *values = distances->values;
   unsigned nbobjs = distances->nbobjs;
   unsigned i, j, k;
 
-  if (!dist || strcmp(dist->name, "NVLinkBandwidth")) {
-    errno = EINVAL;
-    return -1;
-  }
-
   for(i=0; i<nbobjs; i++) {
     hwloc_uint64_t bw_i2sw = 0;
     if (is_nvswitch(objs[i]))
@@ -1475,7 +1461,7 @@ hwloc__distances_transform_transitive_closure(hwloc_topology_t topology,
 }
 
 int
-hwloc_distances_transform(hwloc_topology_t topology,
+hwloc_distances_transform(hwloc_topology_t topology __hwloc_attribute_unused,
                           struct hwloc_distances_s *distances,
                           enum hwloc_distances_transform_e transform,
                           void *transform_attr,
@@ -1494,13 +1480,13 @@ hwloc_distances_transform(hwloc_topology_t topology,
   case HWLOC_DISTANCES_TRANSFORM_MERGE_SWITCH_PORTS:
   {
     int err;
-    err = hwloc__distances_transform_merge_switch_ports(topology, distances);
+    err = hwloc__distances_transform_merge_switch_ports(distances);
     if (!err)
       err = hwloc__distances_transform_remove_null(distances);
     return err;
   }
   case HWLOC_DISTANCES_TRANSFORM_TRANSITIVE_CLOSURE:
-    return hwloc__distances_transform_transitive_closure(topology, distances);
+    return hwloc__distances_transform_transitive_closure(distances);
   default:
     errno = EINVAL;
     return -1;

include/hwloc/distances.h

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2010-2024 Inria.  All rights reserved.
+ * Copyright © 2010-2025 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -234,17 +234,24 @@ enum hwloc_distances_transform_e {
   HWLOC_DISTANCES_TRANSFORM_LINKS = 1,
 
   /** \brief Merge switches with multiple ports into a single object.
-   * This currently only applies to NVSwitches where GPUs seem connected to different
-   * separate switch ports in the NVLinkBandwidth matrix. This transformation will
-   * replace all of them with the same port connected to all GPUs.
-   * Other ports are removed by applying ::HWLOC_DISTANCES_TRANSFORM_REMOVE_NULL internally.
+   *
+   * This currently only applies to NVSwitches where GPUs seem connected
+   * to different switch ports. Switch ports must be objects with subtype
+   * "NVSwitch" as in the NVLinkBandwidth matrix.
+   *
+   * This transformation will replace all ports with only the first one,
+   * now connected to all GPUs. Other ports are removed by applying
+   * ::HWLOC_DISTANCES_TRANSFORM_REMOVE_NULL internally.
    * \hideinitializer
    */
   HWLOC_DISTANCES_TRANSFORM_MERGE_SWITCH_PORTS = 2,
 
   /** \brief Apply a transitive closure to the matrix to connect objects across switches.
-   * This currently only applies to GPUs and NVSwitches in the NVLinkBandwidth matrix.
-   * All pairs of GPUs will be reported as directly connected.
+   *
+   * All pairs of GPUs will be reported as directly connected instead GPUs being
+   * only connected to switches.
+   *
+   * Switch ports must be objects with subtype "NVSwitch" as in the NVLinkBandwidth matrix.
    * \hideinitializer
    */
   HWLOC_DISTANCES_TRANSFORM_TRANSITIVE_CLOSURE = 3