Ecosystem/xmrig
Ecosystem/xmrig
1
0
Fork 0
mirror of https://github.com/xmrig/xmrig.git synced 2024-12-22 19:49:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Update hwloc for MSVC builds.

Browse source
This commit is contained in:
XMRig 2022-03-07 04:29:13 +07:00
parent bbb19ea2f9
commit f25e65b5ac
No known key found for this signature in database
GPG key ID: 446A53638BE94409
19 changed files with 314 additions and 255 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

70
src/3rdparty/hwloc/NEWS vendored
View file

@ -17,6 +17,76 @@ bug fixes (and other actions) for each version of hwloc since version
0.9.
Version 2.7.0
-------------
* Backends
+ Add support for NUMA nodes and caches with more than 64 PUs across
multiple processor groups on Windows 11 and Windows Server 2022.
+ Group objects are not created for Windows processor groups anymore,
except if HWLOC_WINDOWS_PROCESSOR_GROUP_OBJS=1 in the environment.
+ Expose "Cluster" group objects on Linux kernel 5.16+ for CPUs
that share some internal cache or bus. This can be equivalent
to the L2 Cache level on some platforms (e.g. x86) or a specific
level between L2 and L3 on others (e.g. ARM Kungpeng 920).
Thanks to Jonathan Cameron for the help.
- HWLOC_DONT_MERGE_CLUSTER_GROUPS=1 may be set in the environment
to prevent these groups from being merged with identical caches, etc.
+ Improve the oneAPI LevelZero backend:
- Expose subdevices such as "ze0.1" inside root OS devices ("ze0")
when the hardware contains multiple subdevices.
- Add many new attributes to describe device type, and the
numbers of slices, subslices, execution units and threads.
- Expose the memory information as LevelZeroHBM/DDR/MemorySize infos.
+ Ignore the max frequencies of cores in Linux cpukinds when the
base frequencies are available (to avoid exposing hybrid CPUs
when Intel Turbo Boost Max 3.0 gives slightly different max
frequencies to CPU cores).
- May be reverted by setting HWLOC_CPUKINDS_MAXFREQ=1 in the environment.
* Tools
+ Add --grey and --palette options to switch lstopo to greyscale or
white-background-only graphics, or to tune individual colors.
* Build
+ Windows CMake builds now support non-MSVC compilers, detect several
features at build time, can build/run tests, etc.
Thanks to Michael Hirsch and Alexander Neumann .
Version 2.6.0
-------------
* Backends
+ Expose two cpukinds for energy-efficient cores (icestorm) and
high-performance cores (firestorm) on Apple M1 on Mac OS X.
+ Use sysfs CPU "capacity" to rank hybrid cores by efficiency
on Linux when available (mostly on recent ARM platforms for now).
+ Improve HWLOC_MEMBIND_BIND (without the STRICT flag) on Linux kernel
>= 5.15: If more than one node is given, the kernel may now use all
of them instead of only the first one before falling back to others.
+ Expose cache os_index when available on Linux, it may be needed
when using resctrl to configure cache partitioning, memory bandwidth
monitoring, etc.
+ Add a "XGMIHops" distances matrix in the RSMI backend for AMD GPU
interconnected through XGMI links.
+ Expose AMD GPU memory information (VRAM and GTT) in the RSMI backend.
+ Add OS devices such as "bxi0" for Atos/Bull BXI HCAs on Linux.
* Tools
+ lstopo has a better placement algorithm with respect to I/O
objects, see --children-order in the manpage for details.
+ hwloc-annotate may now change object subtypes and cache or memory
sizes.
* Build
+ Allow to specify the ROCm installation for building the RSMI backend:
- Use a custom installation path if specified with --with-rocm=<dir>.
- Use /opt/rocm-<version> if specified with --with-rocm-version=<version>
or the ROCM_VERSION environment variable.
- Try /opt/rocm if it exists.
- See "How do I enable ROCm SMI and select which version to use?"
in the FAQ for details.
+ Add a CMakeLists for Windows under contrib/windows-cmake/ .
* Documentation
+ Add FAQ entry "How do I create a custom heterogeneous and
asymmetric topology?"
Version 2.5.0
-------------
* API

6
src/3rdparty/hwloc/VERSION vendored
View file

@ -8,7 +8,7 @@
# Please update HWLOC_VERSION* in contrib/windows/hwloc_config.h too.
major=2
minor=5
minor=7
release=0
# greek is used for alpha or beta release tags. If it is non-empty,
@ -22,7 +22,7 @@ greek=
# The date when this release was created
date="Jun 14, 2021"
date="Dec 06, 2021"
# If snapshot=1, then use the value from snapshot_version as the
# entire hwloc version (i.e., ignore major, minor, release, and
@ -41,7 +41,7 @@ snapshot_version=${major}.${minor}.${release}${greek}-git
# 2. Version numbers are described in the Libtool current:revision:age
# format.
libhwloc_so_version=20:0:5
libhwloc_so_version=20:2:5
libnetloc_so_version=0:0:0
# Please also update the <TargetName> lines in contrib/windows/libhwloc.vcxproj

45
src/3rdparty/hwloc/include/hwloc.h vendored
View file

@ -346,7 +346,8 @@ typedef enum hwloc_obj_osdev_type_e {
* For instance the "eth0" interface on Linux. */
HWLOC_OBJ_OSDEV_OPENFABRICS, /**< \brief Operating system openfabrics device.
* For instance the "mlx4_0" InfiniBand HCA,
* or "hfi1_0" Omni-Path interface on Linux. */
* "hfi1_0" Omni-Path interface,
* or "bxi0" Atos/Bull BXI HCA on Linux. */
HWLOC_OBJ_OSDEV_DMA, /**< \brief Operating system dma engine device.
* For instance the "dma0chan0" DMA channel on Linux. */
HWLOC_OBJ_OSDEV_COPROC /**< \brief Operating system co-processor device.
@ -1212,8 +1213,9 @@ HWLOC_DECLSPEC int hwloc_set_cpubind(hwloc_topology_t topology, hwloc_const_cpus
/** \brief Get current process or thread binding.
*
* Writes into \p set the physical cpuset which the process or thread (according to \e
* flags) was last bound to.
* The CPU-set \p set (previously allocated by the caller)
* is filled with the list of PUs which the process or
* thread (according to \e flags) was last bound to.
*/
HWLOC_DECLSPEC int hwloc_get_cpubind(hwloc_topology_t topology, hwloc_cpuset_t set, int flags);
@ -1231,6 +1233,10 @@ HWLOC_DECLSPEC int hwloc_get_cpubind(hwloc_topology_t topology, hwloc_cpuset_t s
HWLOC_DECLSPEC int hwloc_set_proc_cpubind(hwloc_topology_t topology, hwloc_pid_t pid, hwloc_const_cpuset_t set, int flags);
/** \brief Get the current physical binding of process \p pid.
*
* The CPU-set \p set (previously allocated by the caller)
* is filled with the list of PUs which the process
* was last bound to.
*
* \note \p hwloc_pid_t is \p pid_t on Unix platforms,
* and \p HANDLE on native Windows platforms.
@ -1256,6 +1262,10 @@ HWLOC_DECLSPEC int hwloc_set_thread_cpubind(hwloc_topology_t topology, hwloc_thr
#ifdef hwloc_thread_t
/** \brief Get the current physical binding of thread \p tid.
*
* The CPU-set \p set (previously allocated by the caller)
* is filled with the list of PUs which the thread
* was last bound to.
*
* \note \p hwloc_thread_t is \p pthread_t on Unix platforms,
* and \p HANDLE on native Windows platforms.
@ -1266,6 +1276,10 @@ HWLOC_DECLSPEC int hwloc_get_thread_cpubind(hwloc_topology_t topology, hwloc_thr
#endif
/** \brief Get the last physical CPU where the current process or thread ran.
*
* The CPU-set \p set (previously allocated by the caller)
* is filled with the list of PUs which the process or
* thread (according to \e flags) last ran on.
*
* The operating system may move some tasks from one processor
* to another at any time according to their binding,
@ -1281,6 +1295,10 @@ HWLOC_DECLSPEC int hwloc_get_thread_cpubind(hwloc_topology_t topology, hwloc_thr
HWLOC_DECLSPEC int hwloc_get_last_cpu_location(hwloc_topology_t topology, hwloc_cpuset_t set, int flags);
/** \brief Get the last physical CPU where a process ran.
*
* The CPU-set \p set (previously allocated by the caller)
* is filled with the list of PUs which the process
* last ran on.
*
* The operating system may move some tasks from one processor
* to another at any time according to their binding,
@ -1511,6 +1529,9 @@ HWLOC_DECLSPEC int hwloc_set_membind(hwloc_topology_t topology, hwloc_const_bitm
/** \brief Query the default memory binding policy and physical locality of the
* current process or thread.
*
* The bitmap \p set (previously allocated by the caller)
* is filled with the process or thread memory binding.
*
* This function has two output parameters: \p set and \p policy.
* The values returned in these parameters depend on both the \p flags
* passed in and the current memory binding policies and nodesets in
@ -1571,6 +1592,9 @@ HWLOC_DECLSPEC int hwloc_set_proc_membind(hwloc_topology_t topology, hwloc_pid_t
/** \brief Query the default memory binding policy and physical locality of the
* specified process.
*
* The bitmap \p set (previously allocated by the caller)
* is filled with the process memory binding.
*
* This function has two output parameters: \p set and \p policy.
* The values returned in these parameters depend on both the \p flags
* passed in and the current memory binding policies and nodesets in
@ -1624,6 +1648,9 @@ HWLOC_DECLSPEC int hwloc_set_area_membind(hwloc_topology_t topology, const void
/** \brief Query the CPUs near the physical NUMA node(s) and binding policy of
* the memory identified by (\p addr, \p len ).
*
* The bitmap \p set (previously allocated by the caller)
* is filled with the memory area binding.
*
* This function has two output parameters: \p set and \p policy.
* The values returned in these parameters depend on both the \p flags
* passed in and the memory binding policies and nodesets of the pages
@ -1652,7 +1679,8 @@ HWLOC_DECLSPEC int hwloc_get_area_membind(hwloc_topology_t topology, const void
/** \brief Get the NUMA nodes where memory identified by (\p addr, \p len ) is physically allocated.
*
* Fills \p set according to the NUMA nodes where the memory area pages
* The bitmap \p set (previously allocated by the caller)
* is filled according to the NUMA nodes where the memory area pages
* are physically allocated. If no page is actually allocated yet,
* \p set may be empty.
*
@ -1698,9 +1726,12 @@ HWLOC_DECLSPEC void *hwloc_alloc_membind(hwloc_topology_t topology, size_t len,
/** \brief Allocate some memory on NUMA memory nodes specified by \p set
*
* This is similar to hwloc_alloc_membind_nodeset() except that it is allowed to change
* the current memory binding policy, thus providing more binding support, at
* the expense of changing the current state.
* First, try to allocate properly with hwloc_alloc_membind().
* On failure, the current process or thread memory binding policy
* is changed with hwloc_set_membind() before allocating memory.
* Thus this function works in more cases, at the expense of changing
* the current state (possibly affecting future allocations that
* would not specify any policy).
*
* If ::HWLOC_MEMBIND_BYNODESET is specified, set is considered a nodeset.
* Otherwise it's a cpuset.

6
src/3rdparty/hwloc/include/hwloc/autogen/config.h vendored
View file

@ -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 © 2009-2011 Cisco Systems, Inc. All rights reserved.
* See COPYING in top-level directory.
@ -11,9 +11,9 @@
#ifndef HWLOC_CONFIG_H
#define HWLOC_CONFIG_H
#define HWLOC_VERSION "2.5.0"
#define HWLOC_VERSION "2.7.0"
#define HWLOC_VERSION_MAJOR 2
#define HWLOC_VERSION_MINOR 5
#define HWLOC_VERSION_MINOR 7
#define HWLOC_VERSION_RELEASE 0
#define HWLOC_VERSION_GREEK ""

25
src/3rdparty/hwloc/include/hwloc/cpukinds.h vendored
View file

@ -1,5 +1,5 @@
/*
* Copyright © 2020 Inria. All rights reserved.
* Copyright © 2020-2021 Inria. All rights reserved.
* See COPYING in top-level directory.
*/
@ -42,18 +42,23 @@ extern "C" {
* (for instance the "CoreType" and "FrequencyMaxMHz",
* see \ref topoattrs_cpukinds).
*
* A higher efficiency value means intrinsic greater performance
* A higher efficiency value means greater intrinsic performance
* (and possibly less performance/power efficiency).
* Kinds with lower efficiency are ranked first:
* Kinds with lower efficiency values are ranked first:
* Passing 0 as \p kind_index to hwloc_cpukinds_get_info() will
* return information about the less efficient CPU kind.
* return information about the CPU kind with lower performance
* but higher energy-efficiency.
* Higher \p kind_index values would rather return information
* about power-hungry high-performance cores.
*
* When available, efficiency values are gathered from the operating
* system (when \p cpukind_efficiency is set in the
* struct hwloc_topology_discovery_support array, only on Windows 10 for now).
* Otherwise hwloc tries to compute efficiencies
* by comparing CPU kinds using frequencies (on ARM),
* or core types and frequencies (on other architectures).
* When available, efficiency values are gathered from the operating system.
* If so, \p cpukind_efficiency is set in the struct hwloc_topology_discovery_support array.
* This is currently available on Windows 10, Mac OS X (Darwin),
* and on some Linux platforms where core "capacity" is exposed in sysfs.
*
* If the operating system does not expose core efficiencies natively,
* hwloc tries to compute efficiencies by comparing CPU kinds using
* frequencies (on ARM), or core types and frequencies (on other architectures).
* The environment variable HWLOC_CPUKINDS_RANKING may be used
* to change this heuristics, see \ref envvar.
*

5
src/3rdparty/hwloc/include/hwloc/distances.h vendored
View file

@ -35,7 +35,8 @@ extern "C" {
* from a core in another node.
* The corresponding kind is ::HWLOC_DISTANCES_KIND_FROM_OS | ::HWLOC_DISTANCES_KIND_FROM_USER.
* The name of this distances structure is "NUMALatency".
* Others distance structures include and "XGMIBandwidth" and "NVLinkBandwidth".
* Others distance structures include and "XGMIBandwidth", "XGMIHops"
* and "NVLinkBandwidth".
*
* The matrix may also contain bandwidths between random sets of objects,
* possibly provided by the user, as specified in the \p kind attribute.
@ -159,7 +160,7 @@ hwloc_distances_get_by_type(hwloc_topology_t topology, hwloc_obj_type_t type,
* Usually only one distances structure may match a given name.
*
* The name of the most common structure is "NUMALatency".
* Others include "XGMIBandwidth" and "NVLinkBandwidth".
* Others include "XGMIBandwidth", "XGMIHops" and "NVLinkBandwidth".
*/
HWLOC_DECLSPEC int
hwloc_distances_get_by_name(hwloc_topology_t topology, const char *name,

9
src/3rdparty/hwloc/include/hwloc/linux.h vendored
View file

@ -1,6 +1,6 @@
/*
* Copyright © 2009 CNRS
* Copyright © 2009-2016 Inria. All rights reserved.
* Copyright © 2009-2021 Inria. All rights reserved.
* Copyright © 2009-2011 Université Bordeaux
* See COPYING in top-level directory.
*/
@ -44,6 +44,10 @@ extern "C" {
HWLOC_DECLSPEC int hwloc_linux_set_tid_cpubind(hwloc_topology_t topology, pid_t tid, hwloc_const_cpuset_t set);
/** \brief Get the current binding of thread \p tid
*
* The CPU-set \p set (previously allocated by the caller)
* is filled with the list of PUs which the thread
* was last bound to.
*
* The behavior is exactly the same as the Linux sched_getaffinity system call,
* but uses a hwloc cpuset.
@ -54,6 +58,9 @@ HWLOC_DECLSPEC int hwloc_linux_set_tid_cpubind(hwloc_topology_t topology, pid_t
HWLOC_DECLSPEC int hwloc_linux_get_tid_cpubind(hwloc_topology_t topology, pid_t tid, hwloc_cpuset_t set);
/** \brief Get the last physical CPU where thread \p tid ran.
*
* The CPU-set \p set (previously allocated by the caller)
* is filled with the PU which the thread last ran on.
*
* \note This is equivalent to calling hwloc_get_proc_last_cpu_location() with
* ::HWLOC_CPUBIND_THREAD as flags.

1
src/3rdparty/hwloc/include/hwloc/plugins.h vendored
View file

@ -497,6 +497,7 @@ hwloc_filter_check_pcidev_subtype_important(unsigned classid)
return (baseclass == 0x03 /* PCI_BASE_CLASS_DISPLAY */
|| baseclass == 0x02 /* PCI_BASE_CLASS_NETWORK */
|| baseclass == 0x01 /* PCI_BASE_CLASS_STORAGE */
|| baseclass == 0x00 /* Unclassified, for Atos/Bull BXI */
|| baseclass == 0x0b /* PCI_BASE_CLASS_PROCESSOR */
|| classid == 0x0c04 /* PCI_CLASS_SERIAL_FIBER */
|| classid == 0x0c06 /* PCI_CLASS_SERIAL_INFINIBAND */

6
src/3rdparty/hwloc/include/private/autogen/config.h vendored
View file

@ -1,6 +1,6 @@
/*
* Copyright © 2009, 2011, 2012 CNRS. All rights reserved.
* Copyright © 2009-2020 Inria. All rights reserved.
* Copyright © 2009-2021 Inria. All rights reserved.
* Copyright © 2009, 2011, 2012, 2015 Université Bordeaux. All rights reserved.
* Copyright © 2009-2020 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
@ -290,10 +290,6 @@
/* Define to '1' if sysctlbyname is present and usable */
/* #undef HAVE_SYSCTLBYNAME */
/* Define to 1 if the system has the type
`SYSTEM_LOGICAL_PROCESSOR_INFORMATION'. */
#define HAVE_SYSTEM_LOGICAL_PROCESSOR_INFORMATION 1
/* Define to 1 if the system has the type
`SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX'. */
#define HAVE_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX 1

2
src/3rdparty/hwloc/include/private/misc.h vendored
View file

@ -504,7 +504,7 @@ hwloc__obj_type_is_icache(hwloc_obj_type_t type)
} \
} while(0)
#else /* HAVE_USELOCALE */
#if __HWLOC_HAVE_ATTRIBUTE_UNUSED
#if HWLOC_HAVE_ATTRIBUTE_UNUSED
#define hwloc_localeswitch_declare int __dummy_nolocale __hwloc_attribute_unused
#define hwloc_localeswitch_init()
#else

1
src/3rdparty/hwloc/include/private/private.h vendored
View file

@ -480,6 +480,7 @@ extern char * hwloc_progname(struct hwloc_topology *topology);
#define HWLOC_GROUP_KIND_AIX_SDL_UNKNOWN 210 /* subkind is SDL level */
#define HWLOC_GROUP_KIND_WINDOWS_PROCESSOR_GROUP 220 /* no subkind */
#define HWLOC_GROUP_KIND_WINDOWS_RELATIONSHIP_UNKNOWN 221 /* no subkind */
#define HWLOC_GROUP_KIND_LINUX_CLUSTER 222 /* no subkind */
/* distance groups */
#define HWLOC_GROUP_KIND_DISTANCE 900 /* subkind is round of adding these groups during distance based grouping */
/* finally, hwloc-specific groups required to insert something else, should disappear as soon as possible */

21
src/3rdparty/hwloc/include/private/windows.h vendored Normal file
View file

@ -0,0 +1,21 @@
/*
* Copyright © 2009 Université Bordeaux
* Copyright © 2020 Inria. All rights reserved.
*
* See COPYING in top-level directory.
*/
#ifndef HWLOC_PRIVATE_WINDOWS_H
#define HWLOC_PRIVATE_WINDOWS_H
#ifdef __GNUC__
#define _ANONYMOUS_UNION __extension__
#define _ANONYMOUS_STRUCT __extension__
#else
#define _ANONYMOUS_UNION
#define _ANONYMOUS_STRUCT
#endif /* __GNUC__ */
#define DUMMYUNIONNAME
#define DUMMYSTRUCTNAME
#endif /* HWLOC_PRIVATE_WINDOWS_H */

7
src/3rdparty/hwloc/src/cpukinds.c vendored
View file

@ -42,6 +42,9 @@ hwloc_internal_cpukinds_dup(hwloc_topology_t new, hwloc_topology_t old)
struct hwloc_internal_cpukind_s *kinds;
unsigned i;
if (!old->nr_cpukinds)
return 0;
kinds = hwloc_tma_malloc(tma, old->nr_cpukinds * sizeof(*kinds));
if (!kinds)
return -1;
@ -445,7 +448,9 @@ static int hwloc__cpukinds_compare_ranking_values(const void *_a, const void *_b
{
const struct hwloc_internal_cpukind_s *a = _a;
const struct hwloc_internal_cpukind_s *b = _b;
return a->ranking_value - b->ranking_value;
uint64_t arv = a->ranking_value;
uint64_t brv = b->ranking_value;
return arv < brv ? -1 : arv > brv ? 1 : 0;
}
/* this function requires ranking values to be unique */

4
src/3rdparty/hwloc/src/memattrs.c vendored
View file

@ -1,5 +1,5 @@
/*
* Copyright © 2020 Inria. All rights reserved.
* Copyright © 2020-2021 Inria. All rights reserved.
* See COPYING in top-level directory.
*/
@ -127,6 +127,8 @@ hwloc_internal_memattrs_dup(struct hwloc_topology *new, struct hwloc_topology *o
struct hwloc_internal_memattr_s *imattrs;
hwloc_memattr_id_t id;
/* old->nr_memattrs is always > 0 thanks to default memattrs */
imattrs = hwloc_tma_malloc(tma, old->nr_memattrs * sizeof(*imattrs));
if (!imattrs)
return -1;

3
src/3rdparty/hwloc/src/pci-common.c vendored
View file

@ -810,13 +810,14 @@ hwloc_pcidisc_find_linkspeed(const unsigned char *config,
* PCIe Gen3 = 8 GT/s signal-rate per lane with 128/130 encoding = 1 GB/s data-rate per lane
* PCIe Gen4 = 16 GT/s signal-rate per lane with 128/130 encoding = 2 GB/s data-rate per lane
* PCIe Gen5 = 32 GT/s signal-rate per lane with 128/130 encoding = 4 GB/s data-rate per lane
* PCIe Gen6 = 64 GT/s signal-rate per lane with 128/130 encoding = 8 GB/s data-rate per lane
*/
/* lanespeed in Gbit/s */
if (speed <= 2)
lanespeed = 2.5f * speed * 0.8f;
else
lanespeed = 8.0f * (1<<(speed-3)) * 128/130; /* assume Gen6 will be 64 GT/s and so on */
lanespeed = 8.0f * (1<<(speed-3)) * 128/130; /* assume Gen7 will be 128 GT/s and so on */
/* linkspeed in GB/s */
*linkspeed = lanespeed * width / 8;

254
src/3rdparty/hwloc/src/topology-windows.c vendored
View file

@ -13,6 +13,7 @@
#include "hwloc.h"
#include "hwloc/windows.h"
#include "private/private.h"
#include "private/windows.h" /* must be before windows.h */
#include "private/debug.h"
#include <windows.h>
@ -65,26 +66,6 @@ typedef enum _LOGICAL_PROCESSOR_RELATIONSHIP {
# endif /* HAVE_RELATIONPROCESSORPACKAGE */
#endif /* HAVE_LOGICAL_PROCESSOR_RELATIONSHIP */
#ifndef HAVE_SYSTEM_LOGICAL_PROCESSOR_INFORMATION
typedef struct _SYSTEM_LOGICAL_PROCESSOR_INFORMATION {
ULONG_PTR ProcessorMask;
LOGICAL_PROCESSOR_RELATIONSHIP Relationship;
_ANONYMOUS_UNION
union {
struct {
BYTE flags;
} ProcessorCore;
struct {
DWORD NodeNumber;
} NumaNode;
CACHE_DESCRIPTOR Cache;
ULONGLONG Reserved[2];
} DUMMYUNIONNAME;
} SYSTEM_LOGICAL_PROCESSOR_INFORMATION, *PSYSTEM_LOGICAL_PROCESSOR_INFORMATION;
#endif
/* Extended interface, for group support */
#ifndef HAVE_GROUP_AFFINITY
typedef struct _GROUP_AFFINITY {
KAFFINITY Mask;
@ -93,35 +74,40 @@ typedef struct _GROUP_AFFINITY {
} GROUP_AFFINITY, *PGROUP_AFFINITY;
#endif
#ifndef HAVE_PROCESSOR_RELATIONSHIP
/* always use our own structure because the EfficiencyClass field didn't exist before Win10 */
typedef struct HWLOC_PROCESSOR_RELATIONSHIP {
BYTE Flags;
BYTE EfficiencyClass; /* for RelationProcessorCore, higher means greater performance but less efficiency, only available in Win10+ */
BYTE EfficiencyClass; /* for RelationProcessorCore, higher means greater performance but less efficiency */
BYTE Reserved[20];
WORD GroupCount;
GROUP_AFFINITY GroupMask[ANYSIZE_ARRAY];
} PROCESSOR_RELATIONSHIP, *PPROCESSOR_RELATIONSHIP;
#endif
} HWLOC_PROCESSOR_RELATIONSHIP;
#ifndef HAVE_NUMA_NODE_RELATIONSHIP
typedef struct _NUMA_NODE_RELATIONSHIP {
/* always use our own structure because the GroupCount and GroupMasks fields didn't exist in some Win10 */
typedef struct HWLOC_NUMA_NODE_RELATIONSHIP {
DWORD NodeNumber;
BYTE Reserved[20];
GROUP_AFFINITY GroupMask;
} NUMA_NODE_RELATIONSHIP, *PNUMA_NODE_RELATIONSHIP;
#endif
BYTE Reserved[18];
WORD GroupCount;
_ANONYMOUS_UNION
union {
GROUP_AFFINITY GroupMask;
GROUP_AFFINITY GroupMasks[ANYSIZE_ARRAY];
} DUMMYUNIONNAME;
} HWLOC_NUMA_NODE_RELATIONSHIP;
#ifndef HAVE_CACHE_RELATIONSHIP
typedef struct _CACHE_RELATIONSHIP {
typedef struct HWLOC_CACHE_RELATIONSHIP {
BYTE Level;
BYTE Associativity;
WORD LineSize;
DWORD CacheSize;
PROCESSOR_CACHE_TYPE Type;
BYTE Reserved[20];
GROUP_AFFINITY GroupMask;
} CACHE_RELATIONSHIP, *PCACHE_RELATIONSHIP;
#endif
BYTE Reserved[18];
WORD GroupCount;
union {
GROUP_AFFINITY GroupMask;
GROUP_AFFINITY GroupMasks[ANYSIZE_ARRAY];
} DUMMYUNIONNAME;
} HWLOC_CACHE_RELATIONSHIP;
#ifndef HAVE_PROCESSOR_GROUP_INFO
typedef struct _PROCESSOR_GROUP_INFO {
@ -141,20 +127,19 @@ typedef struct _GROUP_RELATIONSHIP {
} GROUP_RELATIONSHIP, *PGROUP_RELATIONSHIP;
#endif
#ifndef HAVE_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX
typedef struct _SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX {
/* always use our own structure because we need our own HWLOC_PROCESSOR/CACHE/NUMA_NODE_RELATIONSHIP */
typedef struct HWLOC_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX {
LOGICAL_PROCESSOR_RELATIONSHIP Relationship;
DWORD Size;
_ANONYMOUS_UNION
union {
PROCESSOR_RELATIONSHIP Processor;
NUMA_NODE_RELATIONSHIP NumaNode;
CACHE_RELATIONSHIP Cache;
HWLOC_PROCESSOR_RELATIONSHIP Processor;
HWLOC_NUMA_NODE_RELATIONSHIP NumaNode;
HWLOC_CACHE_RELATIONSHIP Cache;
GROUP_RELATIONSHIP Group;
/* Odd: no member to tell the cpu mask of the package... */
} DUMMYUNIONNAME;
} SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX, *PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX;
#endif
} HWLOC_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX;
#ifndef HAVE_PSAPI_WORKING_SET_EX_BLOCK
typedef union _PSAPI_WORKING_SET_EX_BLOCK {
@ -200,10 +185,7 @@ static PFN_GETCURRENTPROCESSORNUMBER GetCurrentProcessorNumberProc;
typedef VOID (WINAPI *PFN_GETCURRENTPROCESSORNUMBEREX)(PPROCESSOR_NUMBER);
static PFN_GETCURRENTPROCESSORNUMBEREX GetCurrentProcessorNumberExProc;
typedef BOOL (WINAPI *PFN_GETLOGICALPROCESSORINFORMATION)(PSYSTEM_LOGICAL_PROCESSOR_INFORMATION Buffer, PDWORD ReturnLength);
static PFN_GETLOGICALPROCESSORINFORMATION GetLogicalProcessorInformationProc;
typedef BOOL (WINAPI *PFN_GETLOGICALPROCESSORINFORMATIONEX)(LOGICAL_PROCESSOR_RELATIONSHIP relationship, PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX Buffer, PDWORD ReturnLength);
typedef BOOL (WINAPI *PFN_GETLOGICALPROCESSORINFORMATIONEX)(LOGICAL_PROCESSOR_RELATIONSHIP relationship, HWLOC_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *Buffer, PDWORD ReturnLength);
static PFN_GETLOGICALPROCESSORINFORMATIONEX GetLogicalProcessorInformationExProc;
typedef BOOL (WINAPI *PFN_SETTHREADGROUPAFFINITY)(HANDLE hThread, const GROUP_AFFINITY *GroupAffinity, PGROUP_AFFINITY PreviousGroupAffinity);
@ -244,8 +226,6 @@ static void hwloc_win_get_function_ptrs(void)
(PFN_GETACTIVEPROCESSORGROUPCOUNT) GetProcAddress(kernel32, "GetActiveProcessorGroupCount");
GetActiveProcessorCountProc =
(PFN_GETACTIVEPROCESSORCOUNT) GetProcAddress(kernel32, "GetActiveProcessorCount");
GetLogicalProcessorInformationProc =
(PFN_GETLOGICALPROCESSORINFORMATION) GetProcAddress(kernel32, "GetLogicalProcessorInformation");
GetCurrentProcessorNumberProc =
(PFN_GETCURRENTPROCESSORNUMBER) GetProcAddress(kernel32, "GetCurrentProcessorNumber");
GetCurrentProcessorNumberExProc =
@ -370,13 +350,13 @@ static hwloc_cpuset_t * processor_group_cpusets = NULL;
static void
hwloc_win_get_processor_groups(void)
{
PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX procInfoTotal, tmpprocInfoTotal, procInfo;
HWLOC_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *procInfoTotal, *tmpprocInfoTotal, *procInfo;
DWORD length;
unsigned i;
hwloc_debug("querying windows processor groups\n");
if (!GetActiveProcessorGroupCountProc || !GetLogicalProcessorInformationExProc)
if (!GetLogicalProcessorInformationExProc)
goto error;
nr_processor_groups = GetActiveProcessorGroupCountProc();
@ -415,6 +395,8 @@ hwloc_win_get_processor_groups(void)
assert(procInfo->Relationship == RelationGroup);
hwloc_debug("Found %u active windows processor groups\n",
(unsigned) procInfo->Group.ActiveGroupCount);
for (id = 0; id < procInfo->Group.ActiveGroupCount; id++) {
KAFFINITY mask;
hwloc_bitmap_t set;
@ -424,8 +406,8 @@ hwloc_win_get_processor_groups(void)
goto error_with_cpusets;
mask = procInfo->Group.GroupInfo[id].ActiveProcessorMask;
hwloc_debug("group %u %d cpus mask %lx\n", id,
procInfo->Group.GroupInfo[id].ActiveProcessorCount, mask);
hwloc_debug("group %u with %u cpus mask 0x%llx\n", id,
(unsigned) procInfo->Group.GroupInfo[id].ActiveProcessorCount, (unsigned long long) mask);
/* KAFFINITY is ULONG_PTR */
hwloc_bitmap_set_ith_ULONG_PTR(set, id, mask);
/* FIXME: what if running 32bits on a 64bits windows with 64-processor groups?
@ -1008,6 +990,8 @@ hwloc_look_windows(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
unsigned hostname_size = sizeof(hostname);
int has_efficiencyclass = 0;
struct hwloc_win_efficiency_classes eclasses;
char *env = getenv("HWLOC_WINDOWS_PROCESSOR_GROUP_OBJS");
int keep_pgroup_objs = (env && atoi(env));
assert(dstatus->phase == HWLOC_DISC_PHASE_CPU);
@ -1038,137 +1022,8 @@ hwloc_look_windows(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
GetSystemInfo(&SystemInfo);
if (!GetLogicalProcessorInformationExProc && GetLogicalProcessorInformationProc) {
PSYSTEM_LOGICAL_PROCESSOR_INFORMATION procInfo, tmpprocInfo;
unsigned id;
unsigned i;
struct hwloc_obj *obj;
hwloc_obj_type_t type;
length = 0;
procInfo = NULL;
while (1) {
if (GetLogicalProcessorInformationProc(procInfo, &length))
break;
if (GetLastError() != ERROR_INSUFFICIENT_BUFFER)
return -1;
tmpprocInfo = realloc(procInfo, length);
if (!tmpprocInfo) {
free(procInfo);
goto out;
}
procInfo = tmpprocInfo;
}
assert(!length || procInfo);
for (i = 0; i < length / sizeof(*procInfo); i++) {
/* Ignore unknown caches */
if (procInfo->Relationship == RelationCache
&& procInfo->Cache.Type != CacheUnified
&& procInfo->Cache.Type != CacheData
&& procInfo->Cache.Type != CacheInstruction)
continue;
id = HWLOC_UNKNOWN_INDEX;
switch (procInfo[i].Relationship) {
case RelationNumaNode:
type = HWLOC_OBJ_NUMANODE;
id = procInfo[i].NumaNode.NodeNumber;
gotnuma++;
if (id > max_numanode_index)
max_numanode_index = id;
break;
case RelationProcessorPackage:
type = HWLOC_OBJ_PACKAGE;
break;
case RelationCache:
type = (procInfo[i].Cache.Type == CacheInstruction ? HWLOC_OBJ_L1ICACHE : HWLOC_OBJ_L1CACHE) + procInfo[i].Cache.Level - 1;
break;
case RelationProcessorCore:
type = HWLOC_OBJ_CORE;
break;
case RelationGroup:
default:
type = HWLOC_OBJ_GROUP;
break;
}
if (!hwloc_filter_check_keep_object_type(topology, type))
continue;
obj = hwloc_alloc_setup_object(topology, type, id);
obj->cpuset = hwloc_bitmap_alloc();
hwloc_debug("%s#%u mask %llx\n", hwloc_obj_type_string(type), id, (unsigned long long) procInfo[i].ProcessorMask);
/* ProcessorMask is a ULONG_PTR */
hwloc_bitmap_set_ith_ULONG_PTR(obj->cpuset, 0, procInfo[i].ProcessorMask);
hwloc_debug_2args_bitmap("%s#%u bitmap %s\n", hwloc_obj_type_string(type), id, obj->cpuset);
switch (type) {
case HWLOC_OBJ_NUMANODE:
{
ULONGLONG avail;
obj->nodeset = hwloc_bitmap_alloc();
hwloc_bitmap_set(obj->nodeset, id);
if ((GetNumaAvailableMemoryNodeExProc && GetNumaAvailableMemoryNodeExProc(id, &avail))
|| (GetNumaAvailableMemoryNodeProc && GetNumaAvailableMemoryNodeProc(id, &avail))) {
obj->attr->numanode.local_memory = avail;
gotnumamemory++;
}
obj->attr->numanode.page_types_len = 2;
obj->attr->numanode.page_types = malloc(2 * sizeof(*obj->attr->numanode.page_types));
memset(obj->attr->numanode.page_types, 0, 2 * sizeof(*obj->attr->numanode.page_types));
obj->attr->numanode.page_types_len = 1;
obj->attr->numanode.page_types[0].size = SystemInfo.dwPageSize;
#if HAVE_DECL__SC_LARGE_PAGESIZE
obj->attr->numanode.page_types_len++;
obj->attr->numanode.page_types[1].size = sysconf(_SC_LARGE_PAGESIZE);
#endif
break;
}
case HWLOC_OBJ_L1CACHE:
case HWLOC_OBJ_L2CACHE:
case HWLOC_OBJ_L3CACHE:
case HWLOC_OBJ_L4CACHE:
case HWLOC_OBJ_L5CACHE:
case HWLOC_OBJ_L1ICACHE:
case HWLOC_OBJ_L2ICACHE:
case HWLOC_OBJ_L3ICACHE:
obj->attr->cache.size = procInfo[i].Cache.Size;
obj->attr->cache.associativity = procInfo[i].Cache.Associativity == CACHE_FULLY_ASSOCIATIVE ? -1 : procInfo[i].Cache.Associativity ;
obj->attr->cache.linesize = procInfo[i].Cache.LineSize;
obj->attr->cache.depth = procInfo[i].Cache.Level;
switch (procInfo->Cache.Type) {
case CacheUnified:
obj->attr->cache.type = HWLOC_OBJ_CACHE_UNIFIED;
break;
case CacheData:
obj->attr->cache.type = HWLOC_OBJ_CACHE_DATA;
break;
case CacheInstruction:
obj->attr->cache.type = HWLOC_OBJ_CACHE_INSTRUCTION;
break;
default:
hwloc_free_unlinked_object(obj);
continue;
}
break;
case HWLOC_OBJ_GROUP:
obj->attr->group.kind = procInfo[i].Relationship == RelationGroup ? HWLOC_GROUP_KIND_WINDOWS_PROCESSOR_GROUP : HWLOC_GROUP_KIND_WINDOWS_RELATIONSHIP_UNKNOWN;
break;
default:
break;
}
hwloc__insert_object_by_cpuset(topology, NULL, obj, "windows:GetLogicalProcessorInformation");
}
free(procInfo);
}
if (GetLogicalProcessorInformationExProc) {
PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX procInfoTotal, tmpprocInfoTotal, procInfo;
HWLOC_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *procInfoTotal, *tmpprocInfoTotal, *procInfo;
unsigned id;
struct hwloc_obj *obj;
hwloc_obj_type_t type;
@ -1207,8 +1062,16 @@ hwloc_look_windows(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
switch (procInfo->Relationship) {
case RelationNumaNode:
type = HWLOC_OBJ_NUMANODE;
num = 1;
GroupMask = &procInfo->NumaNode.GroupMask;
/* Starting with Windows 11 and Server 2022, the GroupCount field is valid and >=1
* and we may read GroupMasks[]. Older releases have GroupCount==0 and we must read GroupMask.
*/
if (procInfo->NumaNode.GroupCount) {
num = procInfo->NumaNode.GroupCount;
GroupMask = procInfo->NumaNode.GroupMasks;
} else {
num = 1;
GroupMask = &procInfo->NumaNode.GroupMask;
}
id = procInfo->NumaNode.NodeNumber;
gotnuma++;
if (id > max_numanode_index)
@ -1221,18 +1084,20 @@ hwloc_look_windows(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
break;
case RelationCache:
type = (procInfo->Cache.Type == CacheInstruction ? HWLOC_OBJ_L1ICACHE : HWLOC_OBJ_L1CACHE) + procInfo->Cache.Level - 1;
num = 1;
GroupMask = &procInfo->Cache.GroupMask;
/* GroupCount added approximately with NumaNode.GroupCount above */
if (procInfo->Cache.GroupCount) {
num = procInfo->Cache.GroupCount;
GroupMask = procInfo->Cache.GroupMasks;
} else {
num = 1;
GroupMask = &procInfo->Cache.GroupMask;
}
break;
case RelationProcessorCore:
type = HWLOC_OBJ_CORE;
num = procInfo->Processor.GroupCount;
GroupMask = procInfo->Processor.GroupMask;
if (has_efficiencyclass)
/* the EfficiencyClass field didn't exist before Windows10 and recent MSVC headers,
* so just access it manually instead of trying to detect it.
*/
efficiency_class = * ((&procInfo->Processor.Flags) + 1);
efficiency_class = procInfo->Processor.EfficiencyClass;
break;
case RelationGroup:
/* So strange an interface... */
@ -1257,11 +1122,12 @@ hwloc_look_windows(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
groups_pu_set = hwloc_bitmap_alloc();
hwloc_bitmap_or(groups_pu_set, groups_pu_set, set);
if (hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_GROUP)) {
/* Ignore processor groups unless requested and filtered-in */
if (keep_pgroup_objs && hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_GROUP)) {
obj = hwloc_alloc_setup_object(topology, HWLOC_OBJ_GROUP, id);
obj->cpuset = set;
obj->attr->group.kind = HWLOC_GROUP_KIND_WINDOWS_PROCESSOR_GROUP;
hwloc__insert_object_by_cpuset(topology, NULL, obj, "windows:GetLogicalProcessorInformation:ProcessorGroup");
hwloc__insert_object_by_cpuset(topology, NULL, obj, "windows:GetLogicalProcessorInformationEx:ProcessorGroup");
} else
hwloc_bitmap_free(set);
}

37
src/3rdparty/hwloc/src/topology-x86.c vendored
View file

@ -500,7 +500,8 @@ static void read_amd_cores_topoext(struct procinfo *infos, unsigned long flags,
nodes_per_proc = ((ecx >> 8) & 7) + 1;
}
if ((infos->cpufamilynumber == 0x15 && nodes_per_proc > 2)
|| ((infos->cpufamilynumber == 0x17 || infos->cpufamilynumber == 0x18) && nodes_per_proc > 4)) {
|| ((infos->cpufamilynumber == 0x17 || infos->cpufamilynumber == 0x18) && nodes_per_proc > 4)
|| (infos->cpufamilynumber == 0x19 && nodes_per_proc > 1)) {
hwloc_debug("warning: undefined nodes_per_proc value %u, assuming it means %u\n", nodes_per_proc, nodes_per_proc);
}
}
@ -775,13 +776,19 @@ static void look_proc(struct hwloc_backend *backend, struct procinfo *infos, uns
} else if (cpuid_type == amd) {
/* AMD quirks */
if (infos->cpufamilynumber == 0x17
&& cache->level == 3 && cache->nbthreads_sharing == 6) {
/* AMD family 0x17 always shares L3 between 8 APIC ids,
* even when only 6 APIC ids are enabled and reported in nbthreads_sharing
* (on 24-core CPUs).
if (infos->cpufamilynumber >= 0x17 && cache->level == 3) {
/* AMD family 0x19 always shares L3 between 16 APIC ids (8 HT cores).
* while Family 0x17 shares between 8 APIC ids (4 HT cores).
* But many models have less APIC ids enabled and reported in nbthreads_sharing.
* It means we must round-up nbthreads_sharing to the nearest power of 2
* before computing cacheid.
*/
cache->cacheid = infos->apicid / 8;
unsigned nbapics_sharing = cache->nbthreads_sharing;
if (nbapics_sharing & (nbapics_sharing-1))
/* not a power of two, round-up */
nbapics_sharing = 1U<<(1+hwloc_ffsl(nbapics_sharing));
cache->cacheid = infos->apicid / nbapics_sharing;
} else if (infos->cpufamilynumber== 0x10 && infos->cpumodelnumber == 0x9
&& cache->level == 3
@ -807,7 +814,7 @@ static void look_proc(struct hwloc_backend *backend, struct procinfo *infos, uns
} else if (infos->cpufamilynumber == 0x15
&& (infos->cpumodelnumber == 0x1 /* Bulldozer */ || infos->cpumodelnumber == 0x2 /* Piledriver */)
&& cache->level == 3 && cache->nbthreads_sharing == 6) {
/* AMD Bulldozer and Piledriver 12-core processors have same APIC ids as Magny-Cours below,
/* AMD Bulldozer and Piledriver 12-core processors have same APIC ids as Magny-Cours above,
* but we can't merge the checks because the original nbthreads_sharing must be exactly 6 here.
*/
cache->cacheid = (infos->apicid % legacy_max_log_proc) / cache->nbthreads_sharing /* cacheid within the package */
@ -1231,6 +1238,18 @@ static void summarize(struct hwloc_backend *backend, struct procinfo *infos, uns
}
}
cache = hwloc_alloc_setup_object(topology, otype, HWLOC_UNKNOWN_INDEX);
/* We don't specify the os_index of caches because we want to be
* 100% sure they are identical to what the Linux kernel reports
* (so that things like resctrl work).
* However, vendor/model-specific quirks in the x86 code above
* make this difficult.
*
* Caveat: if the x86 backend is used on Linux to avoid kernel bugs,
* IDs won't be available to resctrl users. But resctrl heavily
* relies on the kernel x86 discovery being non-buggy anyway.
*
* TODO: make this optional? or only disable it on Linux?
*/
cache->attr->cache.depth = level;
cache->attr->cache.size = infos[i].cache[l].size;
cache->attr->cache.linesize = infos[i].cache[l].linesize;
@ -1688,7 +1707,7 @@ hwloc_x86_check_cpuiddump_input(const char *src_cpuiddump_path, hwloc_bitmap_t s
char line [32];
dir = opendir(src_cpuiddump_path);
if (!dir)
if (!dir)
return -1;
path = malloc(strlen(src_cpuiddump_path) + strlen("/hwloc-cpuid-info") + 1);

3
src/3rdparty/hwloc/src/topology-xml.c vendored
View file

@ -243,7 +243,7 @@ hwloc__xml_import_object_attr(struct hwloc_topology *topology,
else if (!strcmp(name, "dont_merge")) {
unsigned long lvalue = strtoul(value, NULL, 10);
if (obj->type == HWLOC_OBJ_GROUP)
obj->attr->group.dont_merge = lvalue;
obj->attr->group.dont_merge = (unsigned char) lvalue;
else if (hwloc__xml_verbose())
fprintf(stderr, "%s: ignoring dont_merge attribute for non-group object type\n",
state->global->msgprefix);
@ -2825,6 +2825,7 @@ hwloc__xml_v1export_object_with_memory(hwloc__xml_export_state_t parentstate, hw
/* child has sibling, we must add a Group around those memory children */
hwloc_obj_t group = parentstate->global->v1_memory_group;
parentstate->new_child(parentstate, &gstate, "object");
group->parent = obj->parent;
group->cpuset = obj->cpuset;
group->complete_cpuset = obj->complete_cpuset;
group->nodeset = obj->nodeset;

64
src/3rdparty/hwloc/src/topology.c vendored
View file

@ -69,7 +69,7 @@
* it will break in cygwin, we'll have to use both putenv() and SetEnvironmentVariable().
* Hopefully L0 will be provide a way to enable Sysman without env vars before it happens.
*/
#ifdef HWLOC_HAVE_ATTRIBUTE_CONSTRUCTOR
#if HWLOC_HAVE_ATTRIBUTE_CONSTRUCTOR
static void hwloc_constructor(void) __attribute__((constructor));
static void hwloc_constructor(void)
{
@ -1901,6 +1901,9 @@ hwloc_topology_alloc_group_object(struct hwloc_topology *topology)
static void hwloc_propagate_symmetric_subtree(hwloc_topology_t topology, hwloc_obj_t root);
static void propagate_total_memory(hwloc_obj_t obj);
static void hwloc_set_group_depth(hwloc_topology_t topology);
static void hwloc_connect_children(hwloc_obj_t parent);
static int hwloc_connect_levels(hwloc_topology_t topology);
static int hwloc_connect_special_levels(hwloc_topology_t topology);
hwloc_obj_t
hwloc_topology_insert_group_object(struct hwloc_topology *topology, hwloc_obj_t obj)
@ -2474,13 +2477,26 @@ hwloc_compare_levels_structure(hwloc_topology_t topology, unsigned i)
return 0;
}
/* return > 0 if any level was removed, which means reconnect is needed */
static void
/* return > 0 if any level was removed.
* performs its own reconnect internally if needed
*/
static int
hwloc_filter_levels_keep_structure(hwloc_topology_t topology)
{
unsigned i, j;
int res = 0;
if (topology->modified) {
/* WARNING: hwloc_topology_reconnect() is duplicated partially here
* and at the end of this function:
* - we need normal levels before merging.
* - and we'll need to update special levels after merging.
*/
hwloc_connect_children(topology->levels[0][0]);
if (hwloc_connect_levels(topology) < 0)
return -1;
}
/* start from the bottom since we'll remove intermediate levels */
for(i=topology->nb_levels-1; i>0; i--) {
int replacechild = 0, replaceparent = 0;
@ -2646,6 +2662,22 @@ hwloc_filter_levels_keep_structure(hwloc_topology_t topology)
topology->type_depth[type] = HWLOC_TYPE_DEPTH_MULTIPLE;
}
}
if (res > 0 || topology-> modified) {
/* WARNING: hwloc_topology_reconnect() is duplicated partially here
* and at the beginning of this function.
* If we merged some levels, some child+parent special children lisst
* may have been merged, hence specials level might need reordering,
* So reconnect special levels only here at the end
* (it's not needed at the beginning of this function).
*/
if (hwloc_connect_special_levels(topology) < 0)
return -1;
topology->modified = 0;
}
return 0;
}
static void
@ -2963,9 +2995,9 @@ hwloc_list_special_objects(hwloc_topology_t topology, hwloc_obj_t obj)
}
}
/* Build I/O levels */
/* Build Memory, I/O and Misc levels */
static int
hwloc_connect_io_misc_levels(hwloc_topology_t topology)
hwloc_connect_special_levels(hwloc_topology_t topology)
{
unsigned i;
@ -3176,6 +3208,10 @@ hwloc_connect_levels(hwloc_topology_t topology)
int
hwloc_topology_reconnect(struct hwloc_topology *topology, unsigned long flags)
{
/* WARNING: when updating this function, the replicated code must
* also be updated inside hwloc_filter_levels_keep_structure()
*/
if (flags) {
errno = EINVAL;
return -1;
@ -3188,7 +3224,7 @@ hwloc_topology_reconnect(struct hwloc_topology *topology, unsigned long flags)
if (hwloc_connect_levels(topology) < 0)
return -1;
if (hwloc_connect_io_misc_levels(topology) < 0)
if (hwloc_connect_special_levels(topology) < 0)
return -1;
topology->modified = 0;
@ -3529,15 +3565,12 @@ hwloc_discover(struct hwloc_topology *topology,
}
hwloc_debug_print_objects(0, topology->levels[0][0]);
/* Reconnect things after all these changes.
* Often needed because of Groups inserted for I/Os.
* And required for KEEP_STRUCTURE below.
*/
if (hwloc_topology_reconnect(topology, 0) < 0)
return -1;
hwloc_debug("%s", "\nRemoving levels with HWLOC_TYPE_FILTER_KEEP_STRUCTURE\n");
hwloc_filter_levels_keep_structure(topology);
if (hwloc_filter_levels_keep_structure(topology) < 0)
return -1;
/* takes care of reconnecting children/levels internally,
* because it needs normal levels.
* and it's often needed below because of Groups inserted for I/Os anyway */
hwloc_debug_print_objects(0, topology->levels[0][0]);
/* accumulate children memory in total_memory fields (only once parent is set) */
@ -4360,14 +4393,13 @@ hwloc_topology_restrict(struct hwloc_topology *topology, hwloc_const_bitmap_t se
hwloc_bitmap_free(droppedcpuset);
hwloc_bitmap_free(droppednodeset);
if (hwloc_topology_reconnect(topology, 0) < 0)
if (hwloc_filter_levels_keep_structure(topology) < 0) /* takes care of reconnecting internally */
goto out;
/* some objects may have disappeared, we need to update distances objs arrays */
hwloc_internal_distances_invalidate_cached_objs(topology);
hwloc_internal_memattrs_need_refresh(topology);
hwloc_filter_levels_keep_structure(topology);
hwloc_propagate_symmetric_subtree(topology, topology->levels[0][0]);
propagate_total_memory(topology->levels[0][0]);
hwloc_internal_cpukinds_restrict(topology);