Update hwloc for MSVC.

src/3rdparty/hwloc/include/hwloc.h

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2021 Inria.  All rights reserved.
  * Copyright © 2009-2012 Université Bordeaux
  * Copyright © 2009-2020 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -2362,22 +2362,9 @@ HWLOC_DECLSPEC hwloc_obj_t hwloc_topology_insert_misc_object(hwloc_topology_t to
 /** \brief Allocate a Group object to insert later with hwloc_topology_insert_group_object().
  *
  * This function returns a new Group object.
- * The caller should (at least) initialize its sets before inserting the object.
- * See hwloc_topology_insert_group_object().
  *
- * The \p subtype object attribute may be set to display something else
+ * The caller should (at least) initialize its sets before inserting
- * than "Group" as the type name for this object in lstopo.
+ * the object in the topology. See hwloc_topology_insert_group_object().
- * Custom name/value info pairs may be added with hwloc_obj_add_info() after
- * insertion.
- *
- * The \p kind group attribute should be 0. The \p subkind group attribute may
- * be set to identify multiple Groups of the same level.
- *
- * It is recommended not to set any other object attribute before insertion,
- * since the Group may get discarded during insertion.
- *
- * The object will be destroyed if passed to hwloc_topology_insert_group_object()
- * without any set defined.
  */
 HWLOC_DECLSPEC hwloc_obj_t hwloc_topology_alloc_group_object(hwloc_topology_t topology);
 
@@ -2388,34 +2375,44 @@ HWLOC_DECLSPEC hwloc_obj_t hwloc_topology_alloc_group_object(hwloc_topology_t to
  * the final location of the Group in the topology.
  * Then the object can be passed to this function for actual insertion in the topology.
  *
- * The group \p dont_merge attribute may be set to prevent the core from
- * ever merging this object with another object hierarchically-identical.
- *
  * Either the cpuset or nodeset field (or both, if compatible) must be set
  * to a non-empty bitmap. The complete_cpuset or complete_nodeset may be set
  * instead if inserting with respect to the complete topology
  * (including disallowed, offline or unknown objects).
- *
+ * If grouping several objects, hwloc_obj_add_other_obj_sets() is an easy way
- * It grouping several objects, hwloc_obj_add_other_obj_sets() is an easy way
  * to build the Group sets iteratively.
- *
  * These sets cannot be larger than the current topology, or they would get
  * restricted silently.
- *
  * The core will setup the other sets after actual insertion.
  *
+ * The \p subtype object attribute may be defined (to a dynamically
+ * allocated string) to display something else than "Group" as the
+ * type name for this object in lstopo.
+ * Custom name/value info pairs may be added with hwloc_obj_add_info() after
+ * insertion.
+ *
+ * The group \p dont_merge attribute may be set to \c 1 to prevent
+ * the hwloc core from ever merging this object with another
+ * hierarchically-identical object.
+ * This is useful when the Group itself describes an important feature
+ * that cannot be exposed anywhere else in the hierarchy.
+ *
+ * The group \p kind attribute may be set to a high value such
+ * as \c 0xffffffff to tell hwloc that this new Group should always
+ * be discarded in favor of any existing Group with the same locality.
+ *
  * \return The inserted object if it was properly inserted.
  *
- * \return An existing object if the Group was discarded because the topology already
+ * \return An existing object if the Group was merged or discarded
- * contained an object at the same location (the Group did not add any locality information).
+ * because the topology already contained an object at the same
- * Any name/info key pair set before inserting is appended to the existing object.
+ * location (the Group did not add any hierarchy information).
  *
  * \return \c NULL if the insertion failed because of conflicting sets in topology tree.
  *
  * \return \c NULL if Group objects are filtered-out of the topology (::HWLOC_TYPE_FILTER_KEEP_NONE).
  *
- * \return \c NULL if the object was discarded because no set was initialized in the Group
+ * \return \c NULL if the object was discarded because no set was
- * before insert, or all of them were empty.
+ * initialized in the Group before insert, or all of them were empty.
  */
 HWLOC_DECLSPEC hwloc_obj_t hwloc_topology_insert_group_object(hwloc_topology_t topology, hwloc_obj_t group);
 

src/3rdparty/hwloc/include/hwloc/autogen/config.h

@@ -11,10 +11,10 @@
 #ifndef HWLOC_CONFIG_H
 #define HWLOC_CONFIG_H
 
-#define HWLOC_VERSION "2.4.0"
+#define HWLOC_VERSION "2.4.1"
 #define HWLOC_VERSION_MAJOR 2
 #define HWLOC_VERSION_MINOR 4
-#define HWLOC_VERSION_RELEASE 0
+#define HWLOC_VERSION_RELEASE 1
 #define HWLOC_VERSION_GREEK ""
 
 #define __hwloc_restrict

src/3rdparty/hwloc/include/hwloc/opencl.h

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2012-2020 Inria.  All rights reserved.
+ * Copyright © 2012-2021 Inria.  All rights reserved.
  * Copyright © 2013, 2018 Université Bordeaux.  All right reserved.
  * See COPYING in top-level directory.
  */
@@ -82,9 +82,10 @@ hwloc_opencl_get_device_pci_busid(cl_device_id device,
 	if (CL_SUCCESS == clret
 	    && HWLOC_CL_DEVICE_TOPOLOGY_TYPE_PCIE_AMD == amdtopo.raw.type) {
 		*domain = 0; /* can't do anything better */
-		*bus = (unsigned) amdtopo.pcie.bus;
+		/* cl_device_topology_amd stores bus ID in cl_char, dont convert those signed char directly to unsigned int */
-		*dev = (unsigned) amdtopo.pcie.device;
+		*bus = (unsigned) (unsigned char) amdtopo.pcie.bus;
-		*func = (unsigned) amdtopo.pcie.function;
+		*dev = (unsigned) (unsigned char) amdtopo.pcie.device;
+		*func = (unsigned) (unsigned char) amdtopo.pcie.function;
 		return 0;
 	}
 

src/3rdparty/hwloc/include/hwloc/rsmi.h

@@ -0,0 +1,201 @@
+/*
+ * Copyright © 2012-2020 Inria.  All rights reserved.
+ * Copyright (c) 2020, Advanced Micro Devices, Inc. All rights reserved.
+ * Written by Advanced Micro Devices,
+ * See COPYING in top-level directory.
+ */
+
+/** \file
+ * \brief Macros to help interaction between hwloc and the ROCm SMI Management Library.
+ *
+ * Applications that use both hwloc and the ROCm SMI Management Library may want to
+ * include this file so as to get topology information for AMD GPU devices.
+ */
+
+#ifndef HWLOC_RSMI_H
+#define HWLOC_RSMI_H
+
+#include "hwloc.h"
+#include "hwloc/autogen/config.h"
+#include "hwloc/helper.h"
+#ifdef HWLOC_LINUX_SYS
+#include "hwloc/linux.h"
+#endif
+
+#include <rocm_smi/rocm_smi.h>
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+/** \defgroup hwlocality_rsmi Interoperability with the ROCm SMI Management Library
+ *
+ * This interface offers ways to retrieve topology information about
+ * devices managed by the ROCm SMI Management Library.
+ *
+ * @{
+ */
+
+/** \brief Get the CPU set of logical processors that are physically
+ * close to AMD GPU device whose index is \p dv_ind.
+ *
+ * Return the CPU set describing the locality of the AMD GPU device
+ * whose index is \p dv_ind.
+ *
+ * Topology \p topology and device \p dv_ind must match the local machine.
+ * I/O devices detection and the ROCm SMI component are not needed in the
+ * topology.
+ *
+ * The function only returns the locality of the device.
+ * If more information about the device is needed, OS objects should
+ * be used instead, see hwloc_rsmi_get_device_osdev()
+ * and hwloc_rsmi_get_device_osdev_by_index().
+ *
+ * This function is currently only implemented in a meaningful way for
+ * Linux; other systems will simply get a full cpuset.
+ */
+static __hwloc_inline int
+hwloc_rsmi_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
+                             uint32_t dv_ind, hwloc_cpuset_t set)
+{
+#ifdef HWLOC_LINUX_SYS
+  /* If we're on Linux, use the sysfs mechanism to get the local cpus */
+#define HWLOC_RSMI_DEVICE_SYSFS_PATH_MAX 128
+  char path[HWLOC_RSMI_DEVICE_SYSFS_PATH_MAX];
+  rsmi_status_t ret;
+  uint64_t bdfid = 0;
+  unsigned domain, device, bus;
+
+  if (!hwloc_topology_is_thissystem(topology)) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  ret = rsmi_dev_pci_id_get(dv_ind, &bdfid);
+  if (RSMI_STATUS_SUCCESS != ret) {
+    errno = EINVAL;
+    return -1;
+  }
+  domain = (bdfid>>32) & 0xffffffff;
+  bus = ((bdfid & 0xffff)>>8) & 0xff;
+  device = ((bdfid & 0xff)>>3) & 0x1f;
+
+  sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.0/local_cpus", domain, bus, device);
+  if (hwloc_linux_read_path_as_cpumask(path, set) < 0
+      || hwloc_bitmap_iszero(set))
+    hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
+#else
+  /* Non-Linux systems simply get a full cpuset */
+  hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
+#endif
+  return 0;
+}
+
+/** \brief Get the hwloc OS device object corresponding to the
+ * AMD GPU device whose index is \p dv_ind.
+ *
+ * Return the OS device object describing the AMD GPU device whose
+ * index is \p dv_ind. Returns NULL if there is none.
+ *
+ * The topology \p topology does not necessarily have to match the current
+ * machine. For instance the topology may be an XML import of a remote host.
+ * I/O devices detection and the ROCm SMI component must be enabled in the
+ * topology.
+ *
+ * \note The corresponding PCI device object can be obtained by looking
+ * at the OS device parent object (unless PCI devices are filtered out).
+ */
+static __hwloc_inline hwloc_obj_t
+hwloc_rsmi_get_device_osdev_by_index(hwloc_topology_t topology, uint32_t dv_ind)
+{
+  hwloc_obj_t osdev = NULL;
+  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
+    if (HWLOC_OBJ_OSDEV_GPU == osdev->attr->osdev.type
+      && osdev->name
+      && !strncmp("rsmi", osdev->name, 4)
+      && atoi(osdev->name + 4) == (int) dv_ind)
+      return osdev;
+  }
+  return NULL;
+}
+
+/** \brief Get the hwloc OS device object corresponding to AMD GPU device,
+ * whose index is \p dv_ind.
+ *
+ * Return the hwloc OS device object that describes the given
+ * AMD GPU, whose index is \p dv_ind Return NULL if there is none.
+ *
+ * Topology \p topology and device \p dv_ind must match the local machine.
+ * I/O devices detection and the ROCm SMI component must be enabled in the
+ * topology. If not, the locality of the object may still be found using
+ * hwloc_rsmi_get_device_cpuset().
+ *
+ * \note The corresponding hwloc PCI device may be found by looking
+ * at the result parent pointer (unless PCI devices are filtered out).
+ */
+static __hwloc_inline hwloc_obj_t
+hwloc_rsmi_get_device_osdev(hwloc_topology_t topology, uint32_t dv_ind)
+{
+  hwloc_obj_t osdev;
+  rsmi_status_t ret;
+  uint64_t bdfid = 0;
+  unsigned domain, device, bus, func;
+  uint64_t id;
+  char uuid[64];
+
+  if (!hwloc_topology_is_thissystem(topology)) {
+    errno = EINVAL;
+    return NULL;
+  }
+
+  ret = rsmi_dev_pci_id_get(dv_ind, &bdfid);
+  if (RSMI_STATUS_SUCCESS != ret) {
+    errno = EINVAL;
+    return NULL;
+  }
+  domain = (bdfid>>32) & 0xffffffff;
+  bus = ((bdfid & 0xffff)>>8) & 0xff;
+  device = ((bdfid & 0xff)>>3) & 0x1f;
+  func = bdfid & 0x7;
+
+  ret = rsmi_dev_unique_id_get(dv_ind, &id);
+  if (RSMI_STATUS_SUCCESS != ret)
+    uuid[0] = '\0';
+  else
+    sprintf(uuid, "%lx", id);
+
+  osdev = NULL;
+  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
+    hwloc_obj_t pcidev = osdev->parent;
+    const char *info;
+
+    if (strncmp(osdev->name, "rsmi", 4))
+      continue;
+
+    if (pcidev
+      && pcidev->type == HWLOC_OBJ_PCI_DEVICE
+      && pcidev->attr->pcidev.domain == domain
+      && pcidev->attr->pcidev.bus == bus
+      && pcidev->attr->pcidev.dev == device
+      && pcidev->attr->pcidev.func == func)
+      return osdev;
+
+    info = hwloc_obj_get_info_by_name(osdev, "AMDUUID");
+    if (info && !strcmp(info, uuid))
+      return osdev;
+  }
+
+  return NULL;
+}
+
+/** @} */
+
+
+#ifdef __cplusplus
+} /* extern "C" */
+#endif
+
+
+#endif /* HWLOC_RSMI_H */

src/3rdparty/hwloc/src/cpukinds.c

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2020 Inria.  All rights reserved.
+ * Copyright © 2020-2021 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -270,7 +270,7 @@ hwloc__cpukinds_check_duplicate_rankings(struct hwloc_topology *topology)
   unsigned i,j;
   for(i=0; i<topology->nr_cpukinds; i++)
     for(j=i+1; j<topology->nr_cpukinds; j++)
-      if (topology->cpukinds[i].forced_efficiency == topology->cpukinds[j].forced_efficiency)
+      if (topology->cpukinds[i].ranking_value == topology->cpukinds[j].ranking_value)
         /* if any duplicate, fail */
         return -1;
   return 0;

src/3rdparty/hwloc/src/static-components.h

@@ -1,4 +1,9 @@
-#include <private/internal-components.h>
+HWLOC_DECLSPEC extern const struct hwloc_component hwloc_noos_component;
+HWLOC_DECLSPEC extern const struct hwloc_component hwloc_xml_component;
+HWLOC_DECLSPEC extern const struct hwloc_component hwloc_synthetic_component;
+HWLOC_DECLSPEC extern const struct hwloc_component hwloc_xml_nolibxml_component;
+HWLOC_DECLSPEC extern const struct hwloc_component hwloc_windows_component;
+HWLOC_DECLSPEC extern const struct hwloc_component hwloc_x86_component;
 static const struct hwloc_component * hwloc_static_components[] = {
   &hwloc_noos_component,
   &hwloc_xml_component,

src/3rdparty/hwloc/src/topology-x86.c

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2010-2020 Inria.  All rights reserved.
+ * Copyright © 2010-2021 Inria.  All rights reserved.
  * Copyright © 2010-2013 Université Bordeaux
  * Copyright © 2010-2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -908,6 +908,16 @@ static void summarize(struct hwloc_backend *backend, struct procinfo *infos, uns
   int gotnuma = 0;
   int fulldiscovery = (flags & HWLOC_X86_DISC_FLAG_FULL);
 
+#ifdef HWLOC_DEBUG
+  hwloc_debug("\nSummary of x86 CPUID topology:\n");
+  for(i=0; i<nbprocs; i++) {
+    hwloc_debug("PU %u present=%u apicid=%u on PKG %d CORE %d DIE %d NODE %d\n",
+                i, infos[i].present, infos[i].apicid,
+                infos[i].ids[PKG], infos[i].ids[CORE], infos[i].ids[DIE], infos[i].ids[NODE]);
+  }
+  hwloc_debug("\n");
+#endif
+
   for (i = 0; i < nbprocs; i++)
     if (infos[i].present) {
       hwloc_bitmap_set(complete_cpuset, i);
@@ -1587,7 +1597,8 @@ hwloc_x86_discover(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
   }
 
   if (topology->levels[0][0]->cpuset) {
-    /* somebody else discovered things */
+    /* somebody else discovered things, reconnect levels so that we can look at them */
+    hwloc_topology_reconnect(topology, 0);
     if (topology->nb_levels == 2 && topology->level_nbobjects[1] == data->nbprocs) {
       /* only PUs were discovered, as much as we would, complete the topology with everything else */
       alreadypus = 1;
@@ -1595,7 +1606,6 @@ hwloc_x86_discover(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
     }
 
     /* several object types were added, we can't easily complete, just do partial discovery */
-    hwloc_topology_reconnect(topology, 0);
     ret = hwloc_look_x86(backend, flags);
     if (ret)
       hwloc_obj_add_info(topology->levels[0][0], "Backend", "x86");

src/3rdparty/hwloc/src/topology-xml.c

@@ -1462,6 +1462,9 @@ hwloc__xml_v2import_distances(hwloc_topology_t topology,
 	unsigned long long u;
 	if (heterotypes) {
 	  hwloc_obj_type_t t = HWLOC_OBJ_TYPE_NONE;
+          if (!*tmp)
+            /* reached the end of this indexes attribute */
+            break;
 	  if (hwloc_type_sscanf(tmp, &t, NULL, 0) < 0) {
 	    if (hwloc__xml_verbose())
 	      fprintf(stderr, "%s: %s with unrecognized heterogeneous type %s\n",

src/3rdparty/hwloc/src/topology.c

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2021 Inria.  All rights reserved.
  * Copyright © 2009-2012, 2020 Université Bordeaux
  * Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -567,8 +567,9 @@ hwloc_free_unlinked_object(hwloc_obj_t obj)
 }
 
 /* Replace old with contents of new object, and make new freeable by the caller.
- * Only updates next_sibling/first_child pointers,
+ * Requires reconnect (for siblings pointers and group depth),
- * so may only be used during early discovery.
+ * fixup of sets (only the main cpuset was likely compared before merging),
+ * and update of total_memory and group depth.
  */
 static void
 hwloc_replace_linked_object(hwloc_obj_t old, hwloc_obj_t new)
@@ -1348,7 +1349,7 @@ merge_insert_equal(hwloc_obj_t new, hwloc_obj_t old)
 
 /* returns the result of merge, or NULL if not merged */
 static __hwloc_inline hwloc_obj_t
-hwloc__insert_try_merge_group(hwloc_obj_t old, hwloc_obj_t new)
+hwloc__insert_try_merge_group(hwloc_topology_t topology, hwloc_obj_t old, hwloc_obj_t new)
 {
   if (new->type == HWLOC_OBJ_GROUP && old->type == HWLOC_OBJ_GROUP) {
     /* which group do we keep? */
@@ -1359,6 +1360,7 @@ hwloc__insert_try_merge_group(hwloc_obj_t old, hwloc_obj_t new)
 
       /* keep the new one, it doesn't want to be merged */
       hwloc_replace_linked_object(old, new);
+      topology->modified = 1;
       return new;
 
     } else {
@@ -1366,9 +1368,12 @@ hwloc__insert_try_merge_group(hwloc_obj_t old, hwloc_obj_t new)
 	/* keep the old one, it doesn't want to be merged */
 	return old;
 
-      /* compare subkinds to decice who to keep */
+      /* compare subkinds to decide which group to keep */
-      if (new->attr->group.kind < old->attr->group.kind)
+      if (new->attr->group.kind < old->attr->group.kind) {
+        /* keep smaller kind */
 	hwloc_replace_linked_object(old, new);
+        topology->modified = 1;
+      }
       return old;
     }
   }
@@ -1394,6 +1399,7 @@ hwloc__insert_try_merge_group(hwloc_obj_t old, hwloc_obj_t new)
      * and let the caller free the new object
      */
     hwloc_replace_linked_object(old, new);
+    topology->modified = 1;
     return old;
 
   } else {
@@ -1435,7 +1441,7 @@ hwloc___insert_object_by_cpuset(struct hwloc_topology *topology, hwloc_obj_t cur
     int setres = res;
 
     if (res == HWLOC_OBJ_EQUAL) {
-      hwloc_obj_t merged = hwloc__insert_try_merge_group(child, obj);
+      hwloc_obj_t merged = hwloc__insert_try_merge_group(topology, child, obj);
       if (merged)
 	return merged;
       /* otherwise compare actual types to decide of the inclusion */
@@ -1931,12 +1937,24 @@ hwloc_topology_insert_group_object(struct hwloc_topology *topology, hwloc_obj_t
 
   if (!res)
     return NULL;
-  if (res != obj)
+
-    /* merged */
+  if (res != obj && res->type != HWLOC_OBJ_GROUP)
+    /* merged, not into a Group, nothing to update */
     return res;
 
+  /* res == obj means that the object was inserted.
+   * We need to reconnect levels, fill all its cpu/node sets,
+   * compute its total memory, group depth, etc.
+   *
+   * res != obj usually means that our new group was merged into an
+   * existing object, no need to recompute anything.
+   * However, if merging with an existing group, depending on their kinds,
+   * the contents of obj may overwrite the contents of the old group.
+   * This requires reconnecting levels, filling sets, recomputing total memory, etc.
+   */
+
   /* properly inserted */
-  hwloc_obj_add_children_sets(obj);
+  hwloc_obj_add_children_sets(res);
   if (hwloc_topology_reconnect(topology, 0) < 0)
     return NULL;
 
@@ -1948,7 +1966,7 @@ hwloc_topology_insert_group_object(struct hwloc_topology *topology, hwloc_obj_t
 #endif
     hwloc_topology_check(topology);
 
-  return obj;
+  return res;
 }
 
 hwloc_obj_t
@@ -4658,6 +4676,9 @@ hwloc__check_misc_children(hwloc_topology_t topology, hwloc_bitmap_t gp_indexes,
 static void
 hwloc__check_object(hwloc_topology_t topology, hwloc_bitmap_t gp_indexes, hwloc_obj_t obj)
 {
+  hwloc_uint64_t total_memory;
+  hwloc_obj_t child;
+
   assert(!hwloc_bitmap_isset(gp_indexes, obj->gp_index));
   hwloc_bitmap_set(gp_indexes, obj->gp_index);
 
@@ -4715,6 +4736,18 @@ hwloc__check_object(hwloc_topology_t topology, hwloc_bitmap_t gp_indexes, hwloc_
     assert(hwloc_cache_type_by_depth_type(obj->attr->cache.depth, obj->attr->cache.type) == obj->type);
   }
 
+  /* check total memory */
+  total_memory = 0;
+  if (obj->type == HWLOC_OBJ_NUMANODE)
+    total_memory += obj->attr->numanode.local_memory;
+  for_each_child(child, obj) {
+    total_memory += child->total_memory;
+  }
+  for_each_memory_child(child, obj) {
+    total_memory += child->total_memory;
+  }
+  assert(total_memory == obj->total_memory);
+
   /* check children */
   hwloc__check_normal_children(topology, gp_indexes, obj);
   hwloc__check_memory_children(topology, gp_indexes, obj);