helper.h

/*
 * Copyright © 2009 CNRS
 * Copyright © 2009-2013 Inria.  All rights reserved.
 * Copyright © 2009-2012 Université Bordeaux 1
 * Copyright © 2009-2010 Cisco Systems, Inc.  All rights reserved.
 * See COPYING in top-level directory.
 */

#ifndef HWLOC_HELPER_H
#define HWLOC_HELPER_H

#ifndef HWLOC_H
#error Please include the main hwloc.h instead
#endif

#include <stdlib.h>
#include <errno.h>


#ifdef __cplusplus
extern "C" {
#endif


static __hwloc_inline hwloc_obj_t
hwloc_get_first_largest_obj_inside_cpuset(hwloc_topology_t topology, hwloc_const_cpuset_t set)
{
  hwloc_obj_t obj = hwloc_get_root_obj(topology);
  if (!obj->cpuset || !hwloc_bitmap_intersects(obj->cpuset, set))
    return NULL;
  while (!hwloc_bitmap_isincluded(obj->cpuset, set)) {
    /* while the object intersects without being included, look at its children */
    hwloc_obj_t child = obj->first_child;
    while (child) {
      if (child->cpuset && hwloc_bitmap_intersects(child->cpuset, set))
        break;
      child = child->next_sibling;
    }
    if (!child)
      /* no child intersects, return their father */
      return obj;
    /* found one intersecting child, look at its children */
    obj = child;
  }
  /* obj is included, return it */
  return obj;
}

HWLOC_DECLSPEC int hwloc_get_largest_objs_inside_cpuset (hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                                 hwloc_obj_t * __hwloc_restrict objs, int max);

static __hwloc_inline hwloc_obj_t
hwloc_get_next_obj_inside_cpuset_by_depth (hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                           unsigned depth, hwloc_obj_t prev)
{
  hwloc_obj_t next = hwloc_get_next_obj_by_depth(topology, depth, prev);
  if (!next || !next->cpuset)
    return NULL;
  while (next && !hwloc_bitmap_isincluded(next->cpuset, set))
    next = next->next_cousin;
  return next;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_next_obj_inside_cpuset_by_type (hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                          hwloc_obj_type_t type, hwloc_obj_t prev)
{
  int depth = hwloc_get_type_depth(topology, type);
  if (depth == HWLOC_TYPE_DEPTH_UNKNOWN || depth == HWLOC_TYPE_DEPTH_MULTIPLE)
    return NULL;
  return hwloc_get_next_obj_inside_cpuset_by_depth(topology, set, depth, prev);
}

static __hwloc_inline hwloc_obj_t
hwloc_get_obj_inside_cpuset_by_depth (hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                      unsigned depth, unsigned idx) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_obj_inside_cpuset_by_depth (hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                      unsigned depth, unsigned idx)
{
  hwloc_obj_t obj = hwloc_get_obj_by_depth (topology, depth, 0);
  unsigned count = 0;
  if (!obj || !obj->cpuset)
    return NULL;
  while (obj) {
    if (hwloc_bitmap_isincluded(obj->cpuset, set)) {
      if (count == idx)
        return obj;
      count++;
    }
    obj = obj->next_cousin;
  }
  return NULL;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_obj_inside_cpuset_by_type (hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                     hwloc_obj_type_t type, unsigned idx) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_obj_inside_cpuset_by_type (hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                     hwloc_obj_type_t type, unsigned idx)
{
  int depth = hwloc_get_type_depth(topology, type);
  if (depth == HWLOC_TYPE_DEPTH_UNKNOWN || depth == HWLOC_TYPE_DEPTH_MULTIPLE)
    return NULL;
  return hwloc_get_obj_inside_cpuset_by_depth(topology, set, depth, idx);
}

static __hwloc_inline unsigned
hwloc_get_nbobjs_inside_cpuset_by_depth (hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                         unsigned depth) __hwloc_attribute_pure;
static __hwloc_inline unsigned
hwloc_get_nbobjs_inside_cpuset_by_depth (hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                         unsigned depth)
{
  hwloc_obj_t obj = hwloc_get_obj_by_depth (topology, depth, 0);
  unsigned count = 0;
  if (!obj || !obj->cpuset)
    return 0;
  while (obj) {
    if (hwloc_bitmap_isincluded(obj->cpuset, set))
      count++;
    obj = obj->next_cousin;
  }
  return count;
}

static __hwloc_inline int
hwloc_get_nbobjs_inside_cpuset_by_type (hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                        hwloc_obj_type_t type) __hwloc_attribute_pure;
static __hwloc_inline int
hwloc_get_nbobjs_inside_cpuset_by_type (hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                        hwloc_obj_type_t type)
{
  int depth = hwloc_get_type_depth(topology, type);
  if (depth == HWLOC_TYPE_DEPTH_UNKNOWN)
    return 0;
  if (depth == HWLOC_TYPE_DEPTH_MULTIPLE)
    return -1; /* FIXME: agregate nbobjs from different levels? */
  return hwloc_get_nbobjs_inside_cpuset_by_depth(topology, set, depth);
}

static __hwloc_inline int
hwloc_get_obj_index_inside_cpuset (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_const_cpuset_t set,
                                   hwloc_obj_t obj) __hwloc_attribute_pure;
static __hwloc_inline int
hwloc_get_obj_index_inside_cpuset (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_const_cpuset_t set,
                                   hwloc_obj_t obj)
{
  int idx = 0;
  if (!hwloc_bitmap_isincluded(obj->cpuset, set))
    return -1;
  /* count how many objects are inside the cpuset on the way from us to the beginning of the level */
  while ((obj = obj->prev_cousin) != NULL)
    if (hwloc_bitmap_isincluded(obj->cpuset, set))
      idx++;
  return idx;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_child_covering_cpuset (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_const_cpuset_t set,
                                hwloc_obj_t parent) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_child_covering_cpuset (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_const_cpuset_t set,
                                hwloc_obj_t parent)
{
  hwloc_obj_t child;
  if (!parent->cpuset || hwloc_bitmap_iszero(set))
    return NULL;
  child = parent->first_child;
  while (child) {
    if (child->cpuset && hwloc_bitmap_isincluded(set, child->cpuset))
      return child;
    child = child->next_sibling;
  }
  return NULL;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_obj_covering_cpuset (hwloc_topology_t topology, hwloc_const_cpuset_t set) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_obj_covering_cpuset (hwloc_topology_t topology, hwloc_const_cpuset_t set)
{
  struct hwloc_obj *current = hwloc_get_root_obj(topology);
  if (hwloc_bitmap_iszero(set) || !current->cpuset || !hwloc_bitmap_isincluded(set, current->cpuset))
    return NULL;
  while (1) {
    hwloc_obj_t child = hwloc_get_child_covering_cpuset(topology, set, current);
    if (!child)
      return current;
    current = child;
  }
}

static __hwloc_inline hwloc_obj_t
hwloc_get_next_obj_covering_cpuset_by_depth(hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                            unsigned depth, hwloc_obj_t prev)
{
  hwloc_obj_t next = hwloc_get_next_obj_by_depth(topology, depth, prev);
  if (!next || !next->cpuset)
    return NULL;
  while (next && !hwloc_bitmap_intersects(set, next->cpuset))
    next = next->next_cousin;
  return next;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_next_obj_covering_cpuset_by_type(hwloc_topology_t topology, hwloc_const_cpuset_t set,
                                           hwloc_obj_type_t type, hwloc_obj_t prev)
{
  int depth = hwloc_get_type_depth(topology, type);
  if (depth == HWLOC_TYPE_DEPTH_UNKNOWN || depth == HWLOC_TYPE_DEPTH_MULTIPLE)
    return NULL;
  return hwloc_get_next_obj_covering_cpuset_by_depth(topology, set, depth, prev);
}

static __hwloc_inline hwloc_obj_t
hwloc_get_ancestor_obj_by_depth (hwloc_topology_t topology __hwloc_attribute_unused, unsigned depth, hwloc_obj_t obj) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_ancestor_obj_by_depth (hwloc_topology_t topology __hwloc_attribute_unused, unsigned depth, hwloc_obj_t obj)
{
  hwloc_obj_t ancestor = obj;
  if (obj->depth < depth)
    return NULL;
  while (ancestor && ancestor->depth > depth)
    ancestor = ancestor->parent;
  return ancestor;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_ancestor_obj_by_type (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_obj_type_t type, hwloc_obj_t obj) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_ancestor_obj_by_type (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_obj_type_t type, hwloc_obj_t obj)
{
  hwloc_obj_t ancestor = obj->parent;
  while (ancestor && ancestor->type != type)
    ancestor = ancestor->parent;
  return ancestor;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_common_ancestor_obj (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_obj_t obj1, hwloc_obj_t obj2) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_common_ancestor_obj (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_obj_t obj1, hwloc_obj_t obj2)
{
  /* the loop isn't so easy since intermediate ancestors may have
   * different depth, causing us to alternate between using obj1->parent
   * and obj2->parent. Also, even if at some point we find ancestors of
   * of the same depth, their ancestors may have different depth again.
   */
  while (obj1 != obj2) {
    while (obj1->depth > obj2->depth)
      obj1 = obj1->parent;
    while (obj2->depth > obj1->depth)
      obj2 = obj2->parent;
    if (obj1 != obj2 && obj1->depth == obj2->depth) {
      obj1 = obj1->parent;
      obj2 = obj2->parent;
    }
  }
  return obj1;
}

static __hwloc_inline int
hwloc_obj_is_in_subtree (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_obj_t obj, hwloc_obj_t subtree_root) __hwloc_attribute_pure;
static __hwloc_inline int
hwloc_obj_is_in_subtree (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_obj_t obj, hwloc_obj_t subtree_root)
{
  return hwloc_bitmap_isincluded(obj->cpuset, subtree_root->cpuset);
}

static __hwloc_inline hwloc_obj_t
hwloc_get_next_child (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_obj_t parent, hwloc_obj_t prev)
{
  if (!prev)
    return parent->first_child;
  if (prev->parent != parent)
    return NULL;
  return prev->next_sibling;
}

static __hwloc_inline int
hwloc_get_cache_type_depth (hwloc_topology_t topology,
                            unsigned cachelevel, hwloc_obj_cache_type_t cachetype)
{
  int depth;
  int found = HWLOC_TYPE_DEPTH_UNKNOWN;
  for (depth=0; ; depth++) {
    hwloc_obj_t obj = hwloc_get_obj_by_depth(topology, depth, 0);
    if (!obj)
      break;
    if (obj->type != HWLOC_OBJ_CACHE || obj->attr->cache.depth != cachelevel)
      /* doesn't match, try next depth */
      continue;
    if (cachetype == (hwloc_obj_cache_type_t) -1) {
      if (found != HWLOC_TYPE_DEPTH_UNKNOWN) {
        /* second match, return MULTIPLE */
        return HWLOC_TYPE_DEPTH_MULTIPLE;
      }
      /* first match, mark it as found */
      found = depth;
      continue;
    }
    if (obj->attr->cache.type == cachetype || obj->attr->cache.type == HWLOC_OBJ_CACHE_UNIFIED)
      /* exact match (either unified is alone, or we match instruction or data), return immediately */
      return depth;
  }
  /* went to the bottom, return what we found */
  return found;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_cache_covering_cpuset (hwloc_topology_t topology, hwloc_const_cpuset_t set) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_cache_covering_cpuset (hwloc_topology_t topology, hwloc_const_cpuset_t set)
{
  hwloc_obj_t current = hwloc_get_obj_covering_cpuset(topology, set);
  while (current) {
    if (current->type == HWLOC_OBJ_CACHE)
      return current;
    current = current->parent;
  }
  return NULL;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_shared_cache_covering_obj (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_obj_t obj) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_shared_cache_covering_obj (hwloc_topology_t topology __hwloc_attribute_unused, hwloc_obj_t obj)
{
  hwloc_obj_t current = obj->parent;
  if (!obj->cpuset)
    return NULL;
  while (current && current->cpuset) {
    if (!hwloc_bitmap_isequal(current->cpuset, obj->cpuset)
        && current->type == HWLOC_OBJ_CACHE)
      return current;
    current = current->parent;
  }
  return NULL;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_pu_obj_by_os_index(hwloc_topology_t topology, unsigned os_index) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_pu_obj_by_os_index(hwloc_topology_t topology, unsigned os_index)
{
  hwloc_obj_t obj = NULL;
  while ((obj = hwloc_get_next_obj_by_type(topology, HWLOC_OBJ_PU, obj)) != NULL)
    if (obj->os_index == os_index)
      return obj;
  return NULL;
}

/* TODO: rather provide an iterator? Provide a way to know how much should be allocated? By returning the total number of objects instead? */
HWLOC_DECLSPEC unsigned hwloc_get_closest_objs (hwloc_topology_t topology, hwloc_obj_t src, hwloc_obj_t * __hwloc_restrict objs, unsigned max);

static __hwloc_inline hwloc_obj_t
hwloc_get_obj_below_by_type (hwloc_topology_t topology,
                             hwloc_obj_type_t type1, unsigned idx1,
                             hwloc_obj_type_t type2, unsigned idx2) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_obj_below_by_type (hwloc_topology_t topology,
                             hwloc_obj_type_t type1, unsigned idx1,
                             hwloc_obj_type_t type2, unsigned idx2)
{
  hwloc_obj_t obj;
  obj = hwloc_get_obj_by_type (topology, type1, idx1);
  if (!obj || !obj->cpuset)
    return NULL;
  return hwloc_get_obj_inside_cpuset_by_type(topology, obj->cpuset, type2, idx2);
}

static __hwloc_inline hwloc_obj_t
hwloc_get_obj_below_array_by_type (hwloc_topology_t topology, int nr, hwloc_obj_type_t *typev, unsigned *idxv) __hwloc_attribute_pure;
static __hwloc_inline hwloc_obj_t
hwloc_get_obj_below_array_by_type (hwloc_topology_t topology, int nr, hwloc_obj_type_t *typev, unsigned *idxv)
{
  hwloc_obj_t obj = hwloc_get_root_obj(topology);
  int i;
  for(i=0; i<nr; i++) {
    if (!obj || !obj->cpuset)
      return NULL;
    obj = hwloc_get_obj_inside_cpuset_by_type(topology, obj->cpuset, typev[i], idxv[i]);
  }
  return obj;
}

static __hwloc_inline void
hwloc_distributev(hwloc_topology_t topology, hwloc_obj_t *root, unsigned n_roots, hwloc_cpuset_t *cpuset, unsigned n, unsigned until);
static __hwloc_inline void
hwloc_distribute(hwloc_topology_t topology, hwloc_obj_t root, hwloc_cpuset_t *set, unsigned n, unsigned until)
{
  unsigned i;
  if (!root->arity || n == 1 || root->depth >= until) {
    /* Got to the bottom, we can't split any more, put everything there.  */
    for (i=0; i<n; i++)
      set[i] = hwloc_bitmap_dup(root->cpuset);
    return;
  }
  hwloc_distributev(topology, root->children, root->arity, set, n, until);
}

static __hwloc_inline void
hwloc_distributev(hwloc_topology_t topology, hwloc_obj_t *roots, unsigned n_roots, hwloc_cpuset_t *set, unsigned n, unsigned until)
{
  unsigned i;
  unsigned tot_weight;
  hwloc_cpuset_t *cpusetp = set;

  tot_weight = 0;
  for (i = 0; i < n_roots; i++)
    if (roots[i]->cpuset)
      tot_weight += hwloc_bitmap_weight(roots[i]->cpuset);

  for (i = 0; i < n_roots && tot_weight; i++) {
    /* Give to roots[i] a portion proportional to its weight */
    unsigned weight = roots[i]->cpuset ? hwloc_bitmap_weight(roots[i]->cpuset) : 0;
    unsigned chunk = (n * weight + tot_weight-1) / tot_weight;
    hwloc_distribute(topology, roots[i], cpusetp, chunk, until);
    cpusetp += chunk;
    tot_weight -= weight;
    n -= chunk;
  }
}

static __hwloc_inline hwloc_const_cpuset_t
hwloc_topology_get_complete_cpuset(hwloc_topology_t topology) __hwloc_attribute_pure;
static __hwloc_inline hwloc_const_cpuset_t
hwloc_topology_get_complete_cpuset(hwloc_topology_t topology)
{
  return hwloc_get_root_obj(topology)->complete_cpuset;
}

static __hwloc_inline hwloc_const_cpuset_t
hwloc_topology_get_topology_cpuset(hwloc_topology_t topology) __hwloc_attribute_pure;
static __hwloc_inline hwloc_const_cpuset_t
hwloc_topology_get_topology_cpuset(hwloc_topology_t topology)
{
  return hwloc_get_root_obj(topology)->cpuset;
}

static __hwloc_inline hwloc_const_cpuset_t
hwloc_topology_get_online_cpuset(hwloc_topology_t topology) __hwloc_attribute_pure;
static __hwloc_inline hwloc_const_cpuset_t
hwloc_topology_get_online_cpuset(hwloc_topology_t topology)
{
  return hwloc_get_root_obj(topology)->online_cpuset;
}

static __hwloc_inline hwloc_const_cpuset_t
hwloc_topology_get_allowed_cpuset(hwloc_topology_t topology) __hwloc_attribute_pure;
static __hwloc_inline hwloc_const_cpuset_t
hwloc_topology_get_allowed_cpuset(hwloc_topology_t topology)
{
  return hwloc_get_root_obj(topology)->allowed_cpuset;
}

static __hwloc_inline hwloc_const_nodeset_t
hwloc_topology_get_complete_nodeset(hwloc_topology_t topology) __hwloc_attribute_pure;
static __hwloc_inline hwloc_const_nodeset_t
hwloc_topology_get_complete_nodeset(hwloc_topology_t topology)
{
  return hwloc_get_root_obj(topology)->complete_nodeset;
}

static __hwloc_inline hwloc_const_nodeset_t
hwloc_topology_get_topology_nodeset(hwloc_topology_t topology) __hwloc_attribute_pure;
static __hwloc_inline hwloc_const_nodeset_t
hwloc_topology_get_topology_nodeset(hwloc_topology_t topology)
{
  return hwloc_get_root_obj(topology)->nodeset;
}

static __hwloc_inline hwloc_const_nodeset_t
hwloc_topology_get_allowed_nodeset(hwloc_topology_t topology) __hwloc_attribute_pure;
static __hwloc_inline hwloc_const_nodeset_t
hwloc_topology_get_allowed_nodeset(hwloc_topology_t topology)
{
  return hwloc_get_root_obj(topology)->allowed_nodeset;
}

static __hwloc_inline void
hwloc_cpuset_to_nodeset(hwloc_topology_t topology, hwloc_const_cpuset_t _cpuset, hwloc_nodeset_t nodeset)
{
        int depth = hwloc_get_type_depth(topology, HWLOC_OBJ_NODE);
        hwloc_obj_t obj;

        if (depth == HWLOC_TYPE_DEPTH_UNKNOWN) {
                 if (hwloc_bitmap_iszero(_cpuset))
                        hwloc_bitmap_zero(nodeset);
                else
                        /* Assume the whole system */
                        hwloc_bitmap_fill(nodeset);
                return;
        }

        hwloc_bitmap_zero(nodeset);
        obj = NULL;
        while ((obj = hwloc_get_next_obj_covering_cpuset_by_depth(topology, _cpuset, depth, obj)) != NULL)
                hwloc_bitmap_set(nodeset, obj->os_index);
}

static __hwloc_inline void
hwloc_cpuset_to_nodeset_strict(struct hwloc_topology *topology, hwloc_const_cpuset_t _cpuset, hwloc_nodeset_t nodeset)
{
        int depth = hwloc_get_type_depth(topology, HWLOC_OBJ_NODE);
        hwloc_obj_t obj;
        if (depth == HWLOC_TYPE_DEPTH_UNKNOWN )
                return;
        hwloc_bitmap_zero(nodeset);
        obj = NULL;
        while ((obj = hwloc_get_next_obj_covering_cpuset_by_depth(topology, _cpuset, depth, obj)) != NULL)
                hwloc_bitmap_set(nodeset, obj->os_index);
}

static __hwloc_inline void
hwloc_cpuset_from_nodeset(hwloc_topology_t topology, hwloc_cpuset_t _cpuset, hwloc_const_nodeset_t nodeset)
{
        int depth = hwloc_get_type_depth(topology, HWLOC_OBJ_NODE);
        hwloc_obj_t obj;

        if (depth == HWLOC_TYPE_DEPTH_UNKNOWN ) {
                if (hwloc_bitmap_iszero(nodeset))
                        hwloc_bitmap_zero(_cpuset);
                else
                        /* Assume the whole system */
                        hwloc_bitmap_fill(_cpuset);
                return;
        }

        hwloc_bitmap_zero(_cpuset);
        obj = NULL;
        while ((obj = hwloc_get_next_obj_by_depth(topology, depth, obj)) != NULL) {
                if (hwloc_bitmap_isset(nodeset, obj->os_index))
                        /* no need to check obj->cpuset because objects in levels always have a cpuset */
                        hwloc_bitmap_or(_cpuset, _cpuset, obj->cpuset);
        }
}

static __hwloc_inline void
hwloc_cpuset_from_nodeset_strict(struct hwloc_topology *topology, hwloc_cpuset_t _cpuset, hwloc_const_nodeset_t nodeset)
{
        int depth = hwloc_get_type_depth(topology, HWLOC_OBJ_NODE);
        hwloc_obj_t obj;
        if (depth == HWLOC_TYPE_DEPTH_UNKNOWN )
                return;
        hwloc_bitmap_zero(_cpuset);
        obj = NULL;
        while ((obj = hwloc_get_next_obj_by_depth(topology, depth, obj)) != NULL)
                if (hwloc_bitmap_isset(nodeset, obj->os_index))
                        /* no need to check obj->cpuset because objects in levels always have a cpuset */
                        hwloc_bitmap_or(_cpuset, _cpuset, obj->cpuset);
}

static __hwloc_inline const struct hwloc_distances_s *
hwloc_get_whole_distance_matrix_by_depth(hwloc_topology_t topology, unsigned depth)
{
  hwloc_obj_t root = hwloc_get_root_obj(topology);
  unsigned i;
  for(i=0; i<root->distances_count; i++)
    if (root->distances[i]->relative_depth == depth)
      return root->distances[i];
  return NULL;
}

static __hwloc_inline const struct hwloc_distances_s *
hwloc_get_whole_distance_matrix_by_type(hwloc_topology_t topology, hwloc_obj_type_t type)
{
  int depth = hwloc_get_type_depth(topology, type);
  if (depth < 0)
    return NULL;
  return hwloc_get_whole_distance_matrix_by_depth(topology, depth);
}

static __hwloc_inline const struct hwloc_distances_s *
hwloc_get_distance_matrix_covering_obj_by_depth(hwloc_topology_t topology,
                                                hwloc_obj_t obj, unsigned depth,
                                                unsigned *firstp)
{
  while (obj && obj->cpuset) {
    unsigned i;
    for(i=0; i<obj->distances_count; i++)
      if (obj->distances[i]->relative_depth == depth - obj->depth) {
        if (!obj->distances[i]->nbobjs)
          continue;
        *firstp = hwloc_get_next_obj_inside_cpuset_by_depth(topology, obj->cpuset, depth, NULL)->logical_index;
        return obj->distances[i];
      }
    obj = obj->parent;
  }
  return NULL;
}

static __hwloc_inline int
hwloc_get_latency(hwloc_topology_t topology,
                   hwloc_obj_t obj1, hwloc_obj_t obj2,
                   float *latency, float *reverse_latency)
{
  hwloc_obj_t ancestor;
  const struct hwloc_distances_s * distances;
  unsigned first_logical ;

  if (obj1->depth != obj2->depth) {
    errno = EINVAL;
    return -1;
  }

  ancestor = hwloc_get_common_ancestor_obj(topology, obj1, obj2);
  distances = hwloc_get_distance_matrix_covering_obj_by_depth(topology, ancestor, obj1->depth, &first_logical);
  if (distances && distances->latency) {
    const float * latency_matrix = distances->latency;
    unsigned nbobjs = distances->nbobjs;
    unsigned l1 = obj1->logical_index - first_logical;
    unsigned l2 = obj2->logical_index - first_logical;
    *latency = latency_matrix[l1*nbobjs+l2];
    *reverse_latency = latency_matrix[l2*nbobjs+l1];
    return 0;
  }

  errno = ENOSYS;
  return -1;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_non_io_ancestor_obj(hwloc_topology_t topology __hwloc_attribute_unused,
                              hwloc_obj_t ioobj)
{
  hwloc_obj_t obj = ioobj;
  while (obj && !obj->cpuset) {
    obj = obj->parent;
  }
  return obj;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_next_pcidev(hwloc_topology_t topology, hwloc_obj_t prev)
{
  return hwloc_get_next_obj_by_type(topology, HWLOC_OBJ_PCI_DEVICE, prev);
}

static __hwloc_inline hwloc_obj_t
hwloc_get_pcidev_by_busid(hwloc_topology_t topology,
                          unsigned domain, unsigned bus, unsigned dev, unsigned func)
{
  hwloc_obj_t obj = NULL;
  while ((obj = hwloc_get_next_pcidev(topology, obj)) != NULL) {
    if (obj->attr->pcidev.domain == domain
        && obj->attr->pcidev.bus == bus
        && obj->attr->pcidev.dev == dev
        && obj->attr->pcidev.func == func)
      return obj;
  }
  return NULL;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_pcidev_by_busidstring(hwloc_topology_t topology, const char *busid)
{
  unsigned domain = 0; /* default */
  unsigned bus, dev, func;

  if (sscanf(busid, "%x:%x.%x", &bus, &dev, &func) != 3
      && sscanf(busid, "%x:%x:%x.%x", &domain, &bus, &dev, &func) != 4) {
    errno = EINVAL;
    return NULL;
  }

  return hwloc_get_pcidev_by_busid(topology, domain, bus, dev, func);
}

static __hwloc_inline hwloc_obj_t
hwloc_get_next_osdev(hwloc_topology_t topology, hwloc_obj_t prev)
{
  return hwloc_get_next_obj_by_type(topology, HWLOC_OBJ_OS_DEVICE, prev);
}

static __hwloc_inline hwloc_obj_t
hwloc_get_next_bridge(hwloc_topology_t topology, hwloc_obj_t prev)
{
  return hwloc_get_next_obj_by_type(topology, HWLOC_OBJ_BRIDGE, prev);
}

/* \brief Checks whether a given bridge covers a given PCI bus.
 */
static __hwloc_inline int
hwloc_bridge_covers_pcibus(hwloc_obj_t bridge,
                           unsigned domain, unsigned bus)
{
  return bridge->type == HWLOC_OBJ_BRIDGE
    && bridge->attr->bridge.downstream_type == HWLOC_OBJ_BRIDGE_PCI
    && bridge->attr->bridge.downstream.pci.domain == domain
    && bridge->attr->bridge.downstream.pci.secondary_bus <= bus
    && bridge->attr->bridge.downstream.pci.subordinate_bus >= bus;
}

static __hwloc_inline hwloc_obj_t
hwloc_get_hostbridge_by_pcibus(hwloc_topology_t topology,
                               unsigned domain, unsigned bus)
{
  hwloc_obj_t obj = NULL;
  while ((obj = hwloc_get_next_bridge(topology, obj)) != NULL) {
    if (hwloc_bridge_covers_pcibus(obj, domain, bus)) {
      /* found bridge covering this pcibus, make sure it's a hostbridge */
      assert(obj->attr->bridge.upstream_type == HWLOC_OBJ_BRIDGE_HOST);
      assert(obj->parent->type != HWLOC_OBJ_BRIDGE);
      assert(obj->parent->cpuset);
      return obj;
    }
  }
  return NULL;
}

#ifdef __cplusplus
} /* extern "C" */
#endif


#endif /* HWLOC_HELPER_H */