2 Copyright © 2009-2018 Inria. All rights reserved.
3 Copyright © 2009-2013 Université Bordeaux
4 Copyright © 2009-2011 Cisco Systems, Inc. All rights reserved.
8 Additional copyrights may follow
12 ===========================================================================
14 This file contains the main features as well as overviews of specific
15 bug fixes (and other actions) for each version of hwloc since version
16 0.9 (as initially released as "libtopology", then re-branded to "hwloc"
20 Version 1.11.12 (also included in 2.0.3)
22 * Fix a corner case of hwloc_topology_restrict() where children would
24 * Fix the return length of export_xmlbuffer() functions to always
25 include the ending \0.
28 Version 1.11.11 (also included in 2.0.2)
30 * Add support for Hygon Dhyana processors in the x86 backend,
31 thanks to Pu Wen for the patch.
32 * Fix symbol renaming to also rename internal components,
33 thanks to Evan Ramos for the patch.
34 * Fix build on HP-UX, thanks to Richard Lloyd for reporting the issues.
35 * Detect PCI link speed without being root on Linux >= 4.13.
38 Version 1.11.10 (also included in 2.0.1)
40 * Fix detection of cores and hyperthreads on Mac OS X.
41 * Serialize pciaccess discovery to fix concurrent topology loads in
43 * Fix first touch area memory binding on Linux when thread memory
45 * Some minor fixes to memory binding.
46 * Fix hwloc-dump-hwdata to only process SMBIOS information that correspond
47 to the KNL and KNM configuration.
48 * Add a heuristic for guessing KNL/KNM memory and cluster modes when
49 hwloc-dump-hwdata could not run as root earlier.
50 * Fix discovery of NVMe OS devices on Linux >= 4.0.
51 * Add get_area_memlocation() on Windows.
52 * Add CPUVendor, Model, ... attributes on Mac OS X.
57 * Add support for Zhaoxin ZX-C and ZX-D processors in the x86 backend,
58 thanks to Jeff Zhao for the patch.
59 * Fix AMD Epyc 24-core L3 cache locality in the x86 backend.
60 * Don't crash in the x86 backend when the CPUID vendor string is unknown.
61 * Fix the missing pu discovery support bit on some OS.
62 * Fix the management of the lstopoStyle info attribute for custom colors.
63 * Add verbose warnings when failing to load hwloc v2.0+ XMLs.
68 * Multiple Solaris improvements, thanks to Maureen Chew for the help:
69 + Detect caches on Sparc.
70 + Properly detect allowed/disallowed PUs and NUMA nodes with processor sets.
71 + Add hwloc_get_last_cpu_location() support for the current thread.
72 * Add support for CUDA compute capability 7.0 and fix support for 6.[12].
74 + Fix search for objects by physical index in command-line tools.
75 + Add missing "cpubind:get_thisthread_last_cpu_location" in the output
76 of hwloc-info --support.
77 + Add --pid and --name to specify target processes in hwloc-ps.
78 + Display thread names in lstopo and hwloc-ps on Linux.
80 + Add a FAQ entry about building on Windows.
81 + Install missing sub-manpage for hwloc_obj_add_info() and
82 hwloc_obj_get_info_by_name().
87 * Fix hwloc-bind --membind for CPU-less NUMA nodes (again).
88 Thanks to Gilles Gouaillardet for reporting the issue.
89 * Fix a memory leak on IBM S/390 platforms running Linux.
90 * Fix a memory leak when forcing the x86 backend first on amd64/topoext
91 platforms running Linux.
92 * Command-line tools now support "hbm" instead "numanode" for filtering
93 only high-bandwidth memory nodes when selecting locations.
94 + hwloc-bind also support --hbm and --no-hbm for filtering only or
96 Thanks to Nicolas Denoyelle for the suggestion.
97 * Add --children and --descendants to hwloc-info for listing object
98 children or object descendants of a specific type.
99 * Add --no-index, --index, --no-attrs, --attrs to disable/enable display
100 of index numbers or attributes in the graphical lstopo output.
101 * Try to gather hwloc-dump-hwdata output from all possible locations
102 in hwloc-gather-topology.
103 * Updates to the documentation of locations in hwloc(7) and
104 command-line tools manpages.
109 * Make the Linux discovery about twice faster, especially on the CPU side,
110 by trying to avoid sysfs file accesses as much as possible.
111 * Add support for AMD Family 17h processors (Zen) SMT cores in the Linux
113 * Add the HWLOC_TOPOLOGY_FLAG_THISSYSTEM_ALLOWED_RESOURCES flag (and the
114 HWLOC_THISSYSTEM_ALLOWED_RESOURCES environment variable) for reading the
115 set of allowed resources from the local operating system even if the
116 topology was loaded from XML or synthetic.
117 * Fix hwloc_bitmap_set/clr_range() for infinite ranges that do not
118 overlap currently defined ranges in the bitmap.
119 * Don't reset the lstopo zoom scale when moving the X11 window.
120 * lstopo now has --flags for manually setting topology flags.
121 * hwloc_get_depth_type() returns HWLOC_TYPE_DEPTH_UNKNOWN for Misc objects.
126 * Add support for Knights Mill Xeon Phi, thanks to Piotr Luc for the patch.
127 * Reenable distance gathering on Solaris, disabled by mistake since v1.0.
128 Thanks to TU Wien for the help.
129 * Fix hwloc_get_*obj*_inside_cpuset() functions to ignore objects with
130 empty CPU sets, for instance, CPU-less NUMA nodes such as KNL MCDRAM.
131 Thanks to Nicolas Denoyelle for the report.
132 * Fix XML import of multiple distance matrices.
133 * Add a FAQ entry about "hwloc is only a structural model, it ignores
134 performance models, memory bandwidth, etc.?"
139 * Add MemoryMode and ClusterMode attributes in the Machine object on KNL.
140 Add doc/examples/get-knl-modes.c for an example of retrieving them.
141 Thanks to Grzegorz Andrejczuk.
142 * Fix Linux build with -m32 with respect to libudev.
143 Thanks to Paul Hargrove for reporting the issue.
144 * Fix build with Visual Studio 2015, thanks to Eloi Gaudry for reporting
145 the issue and providing the patch.
146 * Don't forget to display OS device children in the graphical lstopo.
147 * Fix a memory leak on Solaris, thanks to Bryon Gloden for the patch.
148 * Properly handle realloc() failures, thanks to Bryon Gloden for reporting
150 * Fix lstopo crash in ascii/fig/windows outputs when some objects have a
151 lstopoStyle info attribute.
157 + Fix a memory leak on Linux S/390 hosts with books.
158 + Fix /proc/mounts parsing on Linux by using mntent.h.
159 Thanks to Nathan Hjelm for reporting the issue.
160 + Fix a x86 infinite loop on VMware due to the x2APIC feature being
161 advertised without actually being fully supported.
162 Thanks to Jianjun Wen for reporting the problem and testing the patch.
163 + Fix the return value of hwloc_alloc() on mmap() failure.
164 Thanks to Hugo Brunie for reporting the issue.
165 + Fix the return value of command-line tools in some error cases.
166 + Do not break individual thread bindings during x86 backend discovery in a
167 multithreaded process. Thanks to Farouk Mansouri for the report.
168 + Fix hwloc-bind --membind for CPU-less NUMA nodes.
169 + Fix some corner cases in the XML export/import of application userdata.
171 + Add HWLOC_MEMBIND_BYNODESET flag so that membind() functions accept
172 either cpusets or nodesets.
173 + Add hwloc_get_area_memlocation() to check where pages are actually
174 allocated. Only implemented on Linux for now.
175 - There's no _nodeset() variant, but the new flag HWLOC_MEMBIND_BYNODESET
177 + Make hwloc_obj_type_sscanf() parse back everything that may be outputted
178 by hwloc_obj_type_snprintf().
179 * Detection Improvements
180 + Allow the x86 backend to add missing cache levels, so that it completes
181 what the Solaris backend lacks.
182 Thanks to Ryan Zezeski for reporting the issue.
183 + Do not filter-out FibreChannel PCI adapters by default anymore.
184 Thanks to Matt Muggeridge for the report.
185 + Add support for CUDA compute capability 6.x.
187 + Add --support to hwloc-info to list supported features, just like with
188 hwloc_topology_get_support().
189 - Also add --objects and --topology to explicitly switch between the
191 + Add --tid to let hwloc-bind operate on individual threads on Linux.
192 + Add --nodeset to let hwloc-bind report memory binding as NUMA node sets.
193 + hwloc-annotate and lstopo don't drop application userdata from XMLs anymore.
194 - Add --cu to hwloc-annotate to drop these application userdata.
195 + Make the hwloc-dump-hwdata dump directory configurable through configure
196 options such as --runstatedir or --localstatedir.
198 + Add systemd service template contrib/systemd/hwloc-dump-hwdata.service
199 for launching hwloc-dump-hwdata at boot on Linux.
200 Thanks to Grzegorz Andrejczuk.
201 + Add HWLOC_PLUGINS_BLACKLIST environment variable to prevent some plugins
202 from being loaded. Thanks to Alexandre Denis for the suggestion.
203 + Small improvements for various Windows build systems,
204 thanks to Jonathan L Peyton and Marco Atzeri.
209 * Improve support for Intel Knights Landing Xeon Phi on Linux:
210 + Group local NUMA nodes of normal memory (DDR) and high-bandwidth memory
211 (MCDRAM) together through "Cluster" groups so that the local MCDRAM is
213 - See "How do I find the local MCDRAM NUMA node on Intel Knights
214 Landing Xeon Phi?" in the documentation.
215 - For uniformity across all KNL configurations, always have a NUMA node
216 object even if the host is UMA.
217 + Fix the detection of the memory-side cache:
218 - Add the hwloc-dump-hwdata superuser utility to dump SMBIOS information
219 into /var/run/hwloc/ as root during boot, and load this dumped
220 information from the hwloc library at runtime.
221 - See "Why do I need hwloc-dump-hwdata for caches on Intel Knights
222 Landing Xeon Phi?" in the documentation.
223 Thanks to Grzegorz Andrejczuk for the patches and for the help.
224 * The x86 and linux backends may now be combined for discovering CPUs
225 through x86 CPUID and memory from the Linux kernel.
226 This is useful for working around buggy CPU information reported by Linux
227 (for instance the AMD Bulldozer/Piledriver bug below).
228 Combination is enabled by passing HWLOC_COMPONENTS=x86 in the environment.
229 * Fix L3 cache sharing on AMD Opteron 63xx (Piledriver) and 62xx (Bulldozer)
230 in the x86 backend. Thanks to many users who helped.
231 * Fix the overzealous L3 cache sharing fix added to the x86 backend in 1.11.1
232 for AMD Opteron 61xx (Magny-Cours) processors.
233 * The x86 backend may now add the info attribute Inclusive=0 or 1 to caches
234 it discovers, or to caches discovered by other backends earlier.
235 Thanks to Guillaume Beauchamp for the patch.
236 * Fix the management on alloc_membind() allocation failures on AIX, HP-UX
238 * Fix spurious failures to load with ENOMEM on AIX in case of Misc objects
240 * lstopo improvements in X11 and Windows graphical mode:
241 + Add + - f 1 shortcuts to manually zoom-in, zoom-out, reset the scale,
242 or fit the entire window.
243 + Display all keyboard shortcuts in the console.
244 * Debug messages may be disabled at runtime by passing HWLOC_DEBUG_VERBOSE=0
245 in the environment when --enable-debug was passed to configure.
246 * Add a FAQ entry "What are these Group objects in my topology?".
252 + Hardwire the topology of Fujitsu K-computer, FX10, FX100 servers to
253 workaround buggy Linux kernels.
254 Thanks to Takahiro Kawashima and Gilles Gouaillardet.
255 + Fix L3 cache information on AMD Opteron 61xx Magny-Cours processors
256 in the x86 backend. Thanks to Guillaume Beauchamp for the patch.
257 + Detect block devices directly attached to PCI without a controller,
258 for instance NVMe disks. Thanks to Barry M. Tannenbaum.
259 + Add the PCISlot attribute to all PCI functions instead of only the
261 * Miscellaneous internal fixes
262 + Ignore PCI bridges that could fail assertions by reporting buggy
263 secondary-subordinate bus numbers
264 Thanks to George Bosilca for reporting the issue.
265 + Fix an overzealous assertion when inserting an intermediate Group object
266 while Groups are totally ignored.
267 + Fix a memory leak on Linux on AMD processors with dual-core compute units.
268 Thanks to Bob Benner.
269 + Fix a memory leak on failure to load a xml diff file.
270 + Fix some segfaults when inputting an invalid synthetic description.
271 + Fix a segfault when plugins fail to find core symbols.
272 Thanks to Guy Streeter.
273 * Many fixes and improvements in the Windows backend:
274 + Fix the discovery of more than 32 processors and multiple processor
275 groups. Thanks to Barry M. Tannenbaum for the help.
276 + Add thread binding set support in case of multiple process groups.
277 + Add thread binding get support.
278 + Add get_last_cpu_location() support for the current thread.
279 + Disable the unsupported process binding in case of multiple processor
281 + Fix/update the Visual Studio support under contrib/windows.
282 Thanks to Eloi Gaudry for the help.
284 + Fix a segfault when displaying logical indexes in the graphical lstopo.
285 Thanks to Guillaume Mercier for reporting the issue.
286 + Fix lstopo linking with X11 libraries, for instance on Mac OS X.
287 Thanks to Scott Atchley and Pierre Ramet for reporting the issue.
288 + hwloc-annotate, hwloc-diff and hwloc-patch do not drop unavailable
289 resources from the output anymore and those may be annotated as well.
290 + Command-line tools may now import XML from the standard input with -i -.xml
291 + Add missing documentation for the hwloc-info --no-icaches option.
297 + Socket objects are renamed into Package to align with the terminology
298 used by processor vendors. The old HWLOC_OBJ_SOCKET type and "Socket"
299 name are still supported for backward compatibility.
300 + HWLOC_OBJ_NODE is replaced with HWLOC_OBJ_NUMANODE for clarification.
301 HWLOC_OBJ_NODE is still supported for backward compatibility.
302 "Node" and "NUMANode" strings are supported as in earlier releases.
303 * Detection improvements
304 + Add support for Intel Knights Landing Xeon Phi.
305 Thanks to Grzegorz Andrejczuk and Lukasz Anaczkowski.
306 + Add Vendor, Model, Revision, SerialNumber, Type and LinuxDeviceID
307 info attributes to Block OS devices on Linux. Thanks to Vineet Pedaballe
309 - Add --disable-libudev to avoid dependency on the libudev library.
310 + Add "MemoryModule" Misc objects with information about DIMMs, on Linux
311 when privileged and when I/O is enabled.
312 Thanks to Vineet Pedaballe for the help.
313 + Add a PCISlot attribute to PCI devices on Linux when supported to
314 identify the physical PCI slot where the board is plugged.
315 + Add CPUStepping info attribute on x86 processors,
316 thanks to Thomas Röhl for the suggestion.
317 + Ignore the device-tree on non-Power architectures to avoid buggy
318 detection on ARM. Thanks to Orion Poplawski for reporting the issue.
319 + Work-around buggy Xeon E5v3 BIOS reporting invalid PCI-NUMA affinity
320 for the PCI links on the second processor.
321 + Add support for CUDA compute capability 5.x, thanks Benjamin Worpitz.
322 + Many fixes to the x86 backend
323 - Add L1i and fix L2/L3 type on old AMD processors without topoext support.
324 - Fix Intel CPU family and model numbers when basic family isn't 6 or 15.
325 - Fix package IDs on recent AMD processors.
326 - Fix misc issues due to incomplete APIC IDs on x2APIC processors.
327 - Avoid buggy discovery on old SGI Altix UVs with non-unique APIC IDs.
328 + Gather total machine memory on NetBSD.
331 - Collapse identical PCI devices unless --no-collapse is given.
332 This avoids gigantic outputs when a PCI device contains dozens of
333 identical virtual functions.
334 - The ASCII art output is now called "ascii", for instance in
336 The former "txt" extension is retained for backward compatibility.
337 - Automatically scales graphical box width to the inner text in Cairo,
338 ASCII and Windows outputs.
339 - Add --rect to lstopo to force rectangular layout even for NUMA nodes.
340 - Add --restrict-flags to configure the behavior of --restrict.
341 - Objects may have a "Type" info attribute to specify a better type name
342 and display it in lstopo.
343 - Really export all verbose information to the given output file.
345 - May now operate on all types of objects, including I/O.
346 - May now insert Misc objects in the topology.
347 - Do not drop instruction caches and I/O devices from the output anymore.
348 + Fix lstopo path in hwloc-gather-topology after install.
350 + Fix hwloc/cudart.h for machines with multiple PCI domains,
351 thanks to Imre Kerr for reporting the problem.
352 + Fix PCI Bridge-specific depth attribute.
353 + Fix hwloc_bitmap_intersect() for two infinite bitmaps.
354 + Fix some corner cases in the building of levels on large NUMA machines
355 with non-uniform NUMA groups and I/Os.
356 + Improve the performance of object insertion by cpuset for large
358 + Prefix verbose XML import errors with the source name.
359 + Improve pkg-config checks and error messages.
360 + Fix excluding after a component with an argument in the HWLOC_COMPONENTS
361 environment variable.
363 + Fix the recommended way in documentation and examples to allocate memory
364 on some node, it should use HWLOC_MEMBIND_BIND.
365 Thanks to Nicolas Bouzat for reporting the issue.
366 + Add a "Miscellaneous objects" section in the documentation.
367 + Add a FAQ entry "What happens to my topology if I disable symmetric
368 multithreading, hyper-threading, etc. ?" to the documentation.
373 * Actually remove disallowed NUMA nodes from nodesets when the whole-system
375 * Fix the gathering of PCI domains. Thanks to James Custer for reporting
376 the issue and providing a patch.
377 * Fix the merging of identical parent and child in presence of Misc objects.
378 Thanks to Dave Love for reporting the issue.
379 * Fix some misordering of children when merging with ignore_keep_structure()
380 in partially allowed topologies.
381 * Fix an overzealous assertion in the debug code when running on a single-PU
382 host with I/O. Thanks to Thomas Van Doren for reporting the issue.
383 * Don't forget to setup NUMA node object nodesets in x86 backend (for BSDs)
384 and OSF/Tru64 backend.
385 * Fix cpuid-x86 build error with gcc -O3 on x86-32. Thanks to Thomas Van Doren
386 for reporting the issue.
387 * Fix support for future very large caches in the x86 backend.
388 * Fix vendor/device names for SR-IOV PCI devices on Linux.
389 * Fix an unlikely crash in case of buggy hierarchical distance matrix.
390 * Fix PU os_index on some AIX releases. Thanks to Hendryk Bockelmann and
391 Erik Schnetter for helping debugging.
392 * Fix hwloc_bitmap_isincluded() in case of infinite sets.
393 * Change hwloc-ls.desktop into a lstopo.desktop and only install it if
394 lstopo is built with Cairo/X11 support. It cannot work with a non-graphical
396 * Add support for the renaming of Socket into Package in future releases.
397 * Add support for the replacement of HWLOC_OBJ_NODE with HWLOC_OBJ_NUMANODE
399 * Clarify the documentation of distance matrices in hwloc.h and in the manpage
400 of the hwloc-distances. Thanks to Dave Love for the suggestion.
401 * Improve some error messages by displaying more information about the
402 hwloc library in use.
403 * Document how to deal with the ABI break when upgrading to the upcoming 2.0
404 See "How do I handle ABI breaks and API upgrades ?" in the FAQ.
410 + Add hwloc_topology_export_synthetic() to export a topology to a
411 synthetic string without using lstopo. See the Synthetic topologies
412 section in the documentation.
413 + Add hwloc_topology_set/get_userdata() to let the application save
414 a private pointer in the topology whenever it needs a way to find
415 its own object corresponding to a topology.
416 + Add hwloc_get_numanode_obj_by_os_index() and document that this function
417 as well as hwloc_get_pu_obj_by_os_index() are good at converting
418 nodesets and cpusets into objects.
419 + hwloc_distrib() does not ignore any objects anymore when there are
420 too many of them. They get merged with others instead.
421 Thanks to Tim Creech for reporting the issue.
423 + hwloc-bind --get <command-line> now executes the command after displaying
424 the binding instead of ignoring the command entirely.
425 Thanks to John Donners for the suggestion.
426 + Clarify that memory sizes shown in lstopo are local by default
427 unless specified (total memory added in the root object).
428 * Synthetic topologies
429 + Synthetic topology descriptions may now specify attributes such as
430 memory sizes and OS indexes. See the Synthetic topologies section
431 in the documentation.
432 + lstopo now exports in this fully-detailed format by default.
433 The new option --export-synthetic-flags may be used to revert
436 + Add the doc/examples/ subdirectory with several real-life examples,
437 including the already existing hwloc-hello.C for basics.
438 Thanks to Rob Aulwes for the suggestion.
439 + Improve the documentation of CPU and memory binding in the API.
440 + Add a FAQ entry about operating system errors, especially on AMD
441 platforms with buggy cache information.
442 + Add a FAQ entry about loading many topologies in a single program.
444 + Work around buggy Linux kernels reporting 2 sockets instead
445 1 socket with 2 NUMA nodes for each Xeon E5 v3 (Haswell) processor.
446 + pciutils/libpci support is now removed since libpciaccess works
447 well and there's also a Linux-specific PCI backend. For the record,
448 pciutils was GPL and therefore disabled by default since v1.6.2.
449 + Add --disable-cpuid configure flag to work around buggy processor
450 simulators reporting invalid CPUID information.
451 Thanks for Andrew Friedley for reporting the issue.
452 + Fix a racy use of libltdl when manipulating multiple topologies in
454 Thanks to Andra Hugo for reporting the issue and testing patches.
455 + Fix some build failures in private/misc.h.
456 Thanks to Pavan Balaji and Ralph Castain for the reports.
457 + Fix failures to detect X11/Xutil.h on some Solaris platforms.
458 Thanks to Siegmar Gross for reporting the failure.
459 + The plugin ABI has changed, this release will not load plugins
460 built against previous hwloc releases.
465 * Fix a crash when the PCI locality is invalid. Attach to the root object
466 instead. Thanks to Nicolas Denoyelle for reporting the issue.
467 * Fix -f in lstopo manpage. Thanks to Jirka Hladky for reporting the issue.
468 * Fix hwloc_obj_type_sscanf() and others when strncasecmp() is not properly
469 available. Thanks to Nick Papior Andersen for reporting the problem.
470 * Mark Linux file descriptors as close-on-exec to avoid leaks on exec.
471 * Fix some minor memory leaks.
477 + Add hwloc_obj_type_sscanf() to extend hwloc_obj_type_of_string() with
478 type-specific attributes such as Cache/Group depth and Cache type.
479 hwloc_obj_type_of_string() is moved to hwloc/deprecated.h.
480 + Add hwloc_linux_get_tid_last_cpu_location() for retrieving the
481 last CPU where a Linux thread given by TID ran.
482 + Add hwloc_distrib() to extend the old hwloc_distribute[v]() functions.
483 hwloc_distribute[v]() is moved to hwloc/deprecated.h.
484 + Don't mix total and local memory when displaying verbose object attributes
485 with hwloc_obj_attr_snprintf() or in lstopo.
487 + Add CPUVendor, CPUModelNumber and CPUFamilyNumber info attributes for
488 x86, ia64 and Xeon Phi sockets on Linux, to extend the x86-specific
489 support added in v1.8.1. Requested by Ralph Castain.
490 + Add many CPU- and Platform-related info attributes on ARM and POWER
491 platforms, in the Machine and Socket objects.
492 + Add CUDA info attributes describing the number of multiprocessors and
493 cores and the size of the global, shared and L2 cache memories in CUDA
495 + Add OpenCL info attributes describing the number of compute units and
496 the global memory size in OpenCL OS devices.
497 + The synthetic backend now accepts extended types such as L2Cache, L1i or
498 Group3. lstopo also exports synthetic strings using these extended types.
501 - Do not overwrite output files by default anymore.
502 Pass -f or --force to enforce it.
503 - Display OpenCL, CUDA and Xeon Phi numbers of cores and memory sizes
504 in the graphical output.
505 - Fix export to stdout when specifying a Cairo-based output type
508 - Add -e or --get-last-cpu-location to report where processes/threads
509 run instead of where they are bound.
510 - Report locations as likely-more-useful objects such as Cores or Sockets
511 instead of Caches when possible.
513 - Fix failure on Windows when not using --pid.
514 - Add -e as a synonym to --get-last-cpu-location.
516 - Add --reverse to distribute using last objects first and singlify
517 into last bits first. Thanks to Jirka Hladky for the suggestion.
519 - Report unified caches when looking for data or instruction cache
522 + Add experimental Visual Studio support under contrib/windows.
523 Thanks to Eloi Gaudry for his help and for providing the first draft.
524 + Fix some overzealous assertions and warnings about the ordering of
525 objects on a level with respect to cpusets. The ordering is only
526 guaranteed for complete cpusets (based on the first bit in sets).
527 + Fix some memory leaks when importing xml diffs and when exporting a
533 * Fix the cpuid code on Windows 64bits so that the x86 backend gets
534 enabled as expected and can populate CPU information.
535 Thanks to Robin Scher for reporting the problem.
536 * Add CPUVendor/CPUModelNumber/CPUFamilyNumber attributes when running
537 on x86 architecture. Thanks to Ralph Castain for the suggestion.
538 * Work around buggy BIOS reporting duplicate NUMA nodes on Linux.
539 Thanks to Jeff Becker for reporting the problem and testing the patch.
540 * Add a name to the lstopo graphical window. Thanks to Michael Prokop
541 for reporting the issue.
547 + Add the "linuxpci" component that always works on Linux even when
548 libpciaccess and libpci aren't available (and even with a modified
549 file-system root). By default the old "pci" component runs first
550 because "linuxpci" lacks device names (obj->name is always NULL).
552 + Add the topology difference API in hwloc/diff.h for manipulating
553 many similar topologies.
554 + Add hwloc_topology_dup() for duplicating an entire topology.
555 + hwloc.h and hwloc/helper.h have been reorganized to clarify the
556 documentation sections. The actual inline code has moved out of hwloc.h
557 into the new hwloc/inlines.h.
558 + Deprecated functions are now in hwloc/deprecated.h, and not in the
559 official documentation anymore.
561 + Add hwloc-diff and hwloc-patch tools together with the new diff API.
562 + Add hwloc-compress-dir to (de)compress an entire directory of XML files
563 using hwloc-diff and hwloc-patch.
564 + Object colors in the graphical output of lstopo may be changed by adding
565 a "lstopoStyle" info attribute. See CUSTOM COLORS in the lstopo(1) manpage
566 for details. Thanks to Jirka Hladky for discussing the idea.
567 + hwloc-gather-topology may now gather I/O-related files on Linux when
568 --io is given. Only the linuxpci component supports discovering I/O
569 objects from these extended tarballs.
570 + hwloc-annotate now supports --ri to remove/replace info attributes with
572 + hwloc-info supports "root" and "all" special locations for dumping
573 information about the root object.
574 + lstopo now supports --append-legend to append custom lines of text
575 to the legend in the graphical output. Thanks to Jirka Hladky for
577 + hwloc-calc and friends have a more robust parsing of locations given
578 on the command-line and they report useful error messages about it.
579 + Add --whole-system to hwloc-bind, hwloc-calc, hwloc-distances and
580 hwloc-distrib, and add --restrict to hwloc-bind for uniformity among
583 + Calling hwloc_topology_load() or hwloc_topology_set_*() on an already
584 loaded topology now returns an error (deprecated since release 1.6.1).
585 + Fix the initialisation of cpusets and nodesets in Group objects added
586 when inserting PCI hostbridges.
587 + Never merge Group objects that were added explicitly by the user with
588 hwloc_custom_insert_group_object_by_parent().
589 + Add a sanity check during dynamic plugin loading to prevent some
590 crashes when hwloc is dynamically loaded by another plugin mechanisms.
591 + Add --with-hwloc-plugins-path to specify the install/load directories
593 + Add the MICSerialNumber info attribute to the root object when running
594 hwloc inside a Xeon Phi to match the same attribute in the MIC OS device
595 when running in the host.
600 * Do not create invalid block OS devices on very old Linux kernel such
602 * Fix PCI subvendor/device IDs.
603 * Fix the management of Misc objects inserted by parent.
604 Thanks to Jirka Hladky for reporting the problem.
605 * Add a Port<n>State into attribute to OpenFabrics OS devices.
606 * Add a MICSerialNumber info attribute to Xeon PHI/MIC OS devices.
607 * Improve verbose error messages when failing to load from XML.
612 * Fix a failed assertion in the distance grouping code when loading a XML
613 file that already contains some groups.
614 Thanks to Laercio Lima Pilla for reporting the problem.
615 * Remove unexpected Group objects when loading XML topologies with I/O
616 objects and NUMA distances.
617 Thanks to Elena Elkina for reporting the problem and testing patches.
618 * Fix PCI link speed discovery when using libpciaccess.
619 * Fix invalid libpciaccess virtual function device/vendor IDs when using
620 SR-IOV PCI devices on Linux.
621 * Fix GL component build with old NVCtrl releases.
622 Thanks to Jirka Hladky for reporting the problem.
623 * Fix embedding breakage caused by libltdl.
624 Thanks to Pavan Balaji for reporting the problem.
625 * Always use the system-wide libltdl instead of shipping one inside hwloc.
626 * Document issues when enabling plugins while embedding hwloc in another
627 project, in the documentation section Embedding hwloc in Other Software.
628 * Add a FAQ entry "How to get useful topology information on NetBSD?"
629 in the documentation.
630 * Somes fixes in the renaming code for embedding.
631 * Miscellaneous minor build fixes.
636 * New operating system backends
637 + Add BlueGene/Q compute node kernel (CNK) support. See the FAQ in the
638 documentation for details. Thanks to Jeff Hammond, Christopher Samuel
639 and Erik Schnetter for their help.
640 + Add NetBSD support, thanks to Aleksej Saushev.
641 * New I/O device discovery
642 + Add co-processor OS devices such as "mic0" for Intel Xeon Phi (MIC)
643 on Linux. Thanks to Jerome Vienne for helping.
644 + Add co-processor OS devices such as "cuda0" for NVIDIA CUDA-capable GPUs.
645 + Add co-processor OS devices such as "opencl0d0" for OpenCL GPU devices
646 on the AMD OpenCL implementation.
647 + Add GPU OS devices such as ":0.0" for NVIDIA X11 displays.
648 + Add GPU OS devices such as "nvml0" for NVIDIA GPUs.
649 Thanks to Marwan Abdellah and Stefan Eilemann for helping.
650 These new OS devices have some string info attributes such as CoProcType,
651 GPUModel, etc. to better identify them.
652 See the I/O Devices and Attributes documentation sections for details.
654 + Add the "opencl", "cuda", "nvml" and "gl" components for I/O device
656 + "nvml" also improves the discovery of NVIDIA GPU PCIe link speed.
657 All of these new components may be built as plugins. They may also be
658 disabled entirely by passing --disable-opencl/cuda/nvml/gl to configure.
659 See the I/O Devices, Components and Plugins, and FAQ documentation
660 sections for details.
662 + Add hwloc_topology_get_flags().
663 + Add hwloc/plugins.h for building external plugins.
664 See the Adding new discovery components and plugins section.
666 + Add hwloc/opencl.h, hwloc/nvml.h, hwloc/gl.h and hwloc/intel-mic.h
667 to retrieve the locality of OS devices that correspond to AMD OpenCL
668 GPU devices or indexes, to NVML devices or indexes, to NVIDIA X11
669 displays, or to Intel Xeon Phi (MIC) device indexes.
670 + Add new helpers in hwloc/cuda.h and hwloc/cudart.h to convert
671 between CUDA devices or indexes and hwloc OS devices.
672 + Add hwloc_ibv_get_device_osdev() and clarify the requirements
673 of the OpenFabrics Verbs helpers in hwloc/openfabrics-verbs.h.
675 + hwloc-info is not only a synonym of lstopo -s anymore, it also
676 dumps information about objects given on the command-line.
678 + Add a section "Existing components and plugins".
679 + Add a list of common OS devices in section "Software devices".
680 + Add a new FAQ entry "Why is lstopo slow?" about lstopo slowness
681 issues because of GPUs.
682 + Clarify the documentation of inline helpers in hwloc/myriexpress.h
683 and hwloc/openfabrics-verbs.h.
685 + Improve cache detection on AIX.
686 + The HWLOC_COMPONENTS variable now excludes the components whose
687 names are prefixed with '-'.
688 + lstopo --ignore PU now works when displaying the topology in
689 graphical and textual mode (not when exporting to XML).
690 + Make sure I/O options always appear in lstopo usage, not only when
691 using pciutils/libpci.
692 + Remove some unneeded Linux specific includes from some interoperability
694 + Fix some inconsistencies in hwloc-distrib and hwloc-assembler-remote
695 manpages. Thanks to Guy Streeter for the report.
696 + Fix a memory leak on AIX when getting memory binding.
697 + Fix many small memory leaks on Linux.
698 + The `libpci' component is now called `pci' but the old name is still
699 accepted in the HWLOC_COMPONENTS variable for backward compatibility.
704 * Use libpciaccess instead of pciutils/libpci by default for I/O discovery.
705 pciutils/libpci is only used if --enable-libpci is given to configure
706 because its GPL license may taint hwloc. See the Installation section
707 in the documentation for details.
708 * Fix get_cpubind on Solaris when bound to a single PU with
709 processor_bind(). Thanks to Eugene Loh for reporting the problem
710 and providing a patch.
715 * Fix some crash or buggy detection in the x86 backend when Linux
716 cgroups/cpusets restrict the available CPUs.
717 * Fix the pkg-config output with --libs --static.
718 Thanks to Erik Schnetter for reporting one of the problems.
719 * Fix the output of hwloc-calc -H --hierarchical when using logical
720 indexes in the output.
721 * Calling hwloc_topology_load() multiple times on the same topology
722 is officially deprecated. hwloc will warn in such cases.
723 * Add some documentation about existing plugins/components, package
724 dependencies, and I/O devices specification on the command-line.
730 + Reorganize the backend infrastructure to support dynamic selection
731 of components and dynamic loading of plugins. For details, see the
732 new documentation section Components and plugins.
733 - The HWLOC_COMPONENTS variable lets one replace the default discovery
735 - Dynamic loading of plugins may be enabled with --enable-plugins
736 (except on AIX and Windows). It will build libxml2 and libpci
737 support as separated modules. This helps reducing the dependencies
738 of the core hwloc library when distributed as a binary package.
740 + Add CPUModel detection on Darwin and x86/FreeBSD.
741 Thanks to Robin Scher for providing ways to implement this.
742 + The x86 backend now adds CPUModel info attributes to socket objects
743 created by other backends that do not natively support this attribute.
744 + Fix detection on FreeBSD in case of cpuset restriction. Thanks to
745 Sebastian Kuzminsky for reporting the problem.
747 + Add hwloc_topology_set_userdata_import/export_callback(),
748 hwloc_export_obj_userdata() and _userdata_base64() to let
749 applications specify how to save/restore the custom data they placed
750 in the userdata private pointer field of hwloc objects.
752 + Add hwloc-annotate program to add string info attributes to XML
754 + Add --pid-cmd to hwloc-ps to append the output of a command to each
755 PID line. May be used for showing Open MPI process ranks, see the
756 hwloc-ps(1) manpage for details.
757 + hwloc-bind now exits with an error if binding fails; the executable
758 is not launched unless binding suceeeded or --force was given.
759 + Add --quiet to hwloc-calc and hwloc-bind to hide non-fatal error
761 + Fix command-line pid support in windows tools.
762 + All programs accept --verbose as a synonym to -v.
764 + Fix some DIR descriptor leaks on Linux.
765 + Fix I/O device lists when some were filtered out after a XML import.
766 + Fix the removal of I/O objects when importing a I/O-enabled XML topology
767 without any I/O topology flag.
768 + When merging objects with HWLOC_IGNORE_TYPE_KEEP_STRUCTURE or
769 lstopo --merge, compare object types before deciding which one of two
770 identical object to remove (e.g. keep sockets in favor of caches).
771 + Add some GUID- and LID-related info attributes to OpenFabrics
773 + Only add CPUType socket attributes on Solaris/Sparc. Other cases
774 don't report reliable information (Solaris/x86), and a replacement
775 is available as the Architecture string info in the Machine object.
776 + Add missing Backend string info on Solaris in most cases.
777 + Document object attributes and string infos in a new Attributes
778 section in the documentation.
779 + Add a section about Synthetic topologies in the documentation.
782 Version 1.5.2 (some of these changes are in v1.6.2 but not in v1.6)
784 * Use libpciaccess instead of pciutils/libpci by default for I/O discovery.
785 pciutils/libpci is only used if --enable-libpci is given to configure
786 because its GPL license may taint hwloc. See the Installation section
787 in the documentation for details.
788 * Fix get_cpubind on Solaris when bound to a single PU with
789 processor_bind(). Thanks to Eugene Loh for reporting the problem
790 and providing a patch.
791 * Fix some DIR descriptor leaks on Linux.
792 * Fix I/O device lists when some were filtered out after a XML import.
793 * Add missing Backend string info on Solaris in most cases.
794 * Fix the removal of I/O objects when importing a I/O-enabled XML topology
795 without any I/O topology flag.
796 * Fix the output of hwloc-calc -H --hierarchical when using logical
797 indexes in the output.
798 * Fix the pkg-config output with --libs --static.
799 Thanks to Erik Schnetter for reporting one of the problems.
804 * Fix block OS device detection on Linux kernel 3.3 and later.
805 Thanks to Guy Streeter for reporting the problem and testing the fix.
806 * Fix the cpuid code in the x86 backend (for FreeBSD). Thanks to
807 Sebastian Kuzminsky for reporting problems and testing patches.
808 * Fix 64bit detection on FreeBSD.
809 * Fix some corner cases in the management of the thissystem flag with
810 respect to topology flags and environment variables.
811 * Fix some corner cases in command-line parsing checks in hwloc-distrib
813 * Make sure we do not miss some block OS devices on old Linux kernels
814 when a single PCI device has multiple IDE hosts/devices behind it.
815 * Do not disable I/O devices or instruction caches in hwloc-assembler output.
821 + Do not limit the number of processors to 1024 on Solaris anymore.
822 + Gather total machine memory on FreeBSD. Thanks to Cyril Roelandt.
823 + XML topology files do not depend on the locale anymore. Float numbers
824 such as NUMA distances or PCI link speeds now always use a dot as a
826 + Add instruction caches detection on Linux, AIX, Windows and Darwin.
827 + Add get_last_cpu_location() support for the current thread on AIX.
828 + Support binding on AIX when threads or processes were bound with
829 bindprocessor(). Thanks to Hendryk Bockelmann for reporting the issue
830 and testing patches, and to Farid Parpia for explaining the binding
832 + Improve AMD topology detection in the x86 backend (for FreeBSD) using
835 + Increase HWLOC_API_VERSION to 0x00010500 so that API changes may be
836 detected at build-time.
837 + Add a cache type attribute describind Data, Instruction and Unified
838 caches. Caches with different types but same depth (for instance L1d
839 and L1i) are placed on different levels.
840 + Add hwloc_get_cache_type_depth() to retrieve the hwloc level depth of
841 of the given cache depth and type, for instance L1i or L2.
842 It helps disambiguating the case where hwloc_get_type_depth() returns
843 HWLOC_TYPE_DEPTH_MULTIPLE.
844 + Instruction caches are ignored unless HWLOC_TOPOLOGY_FLAG_ICACHES is
845 passed to hwloc_topology_set_flags() before load.
846 + Add hwloc_ibv_get_device_osdev_by_name() OpenFabrics helper in
847 openfabrics-verbs.h to find the hwloc OS device object corresponding to
848 an OpenFabrics device.
850 + Add lstopo-no-graphics, a lstopo built without graphical support to
851 avoid dependencies on external libraries such as Cairo and X11. When
852 supported, graphical outputs are only available in the original lstopo
854 - Packagers splitting lstopo and lstopo-no-graphics into different
855 packages are advised to use the alternatives system so that lstopo
856 points to the best available binary.
857 + Instruction caches are enabled in lstopo by default. Use --no-icaches
859 + Add -t/--threads to show threads in hwloc-ps.
860 * Removal of obsolete components
861 + Remove the old cpuset interface (hwloc/cpuset.h) which is deprecated and
862 superseded by the bitmap API (hwloc/bitmap.h) since v1.1.
863 hwloc_cpuset and nodeset types are still defined, but all hwloc_cpuset_*
864 compatibility wrappers are now gone.
865 + Remove Linux libnuma conversion helpers for the deprecated and
866 broken nodemask_t interface.
867 + Remove support for "Proc" type name, it was superseded by "PU" in v1.0.
868 + Remove hwloc-mask symlinks, it was replaced by hwloc-calc in v1.0.
870 + Fix PCIe 3.0 link speed computation.
871 + Non-printable characters are dropped from strings during XML export.
872 + Fix importing of escaped characters with the minimalistic XML backend.
873 + Assert hwloc_is_thissystem() in several I/O related helpers.
874 + Fix some memory leaks in the x86 backend for FreeBSD.
875 + Minor fixes to ease native builds on Windows.
876 + Limit the number of retries when operating on all threads within a
877 process on Linux if the list of threads is heavily getting modified.
882 * This release is only meant to fix the pciutils license issue when upgrading
883 to hwloc v1.5 or later is not possible. It contains several other minor
884 fixes but ignores many of them that are only in v1.5 or later.
885 * Use libpciaccess instead of pciutils/libpci by default for I/O discovery.
886 pciutils/libpci is only used if --enable-libpci is given to configure
887 because its GPL license may taint hwloc. See the Installation section
888 in the documentation for details.
889 * Fix PCIe 3.0 link speed computation.
890 * Fix importing of escaped characters with the minimalistic XML backend.
891 * Fix a memory leak in the x86 backend.
896 * Fix build on Solaris 9 and earlier when fabsf() is not a compiler
897 built-in. Thanks to Igor Galić for reporting the problem.
898 * Fix support for more than 32 processors on Windows. Thanks to Hartmut
899 Kaiser for reporting the problem.
900 * Fix process-wide binding and cpulocation routines on Linux when some
901 threads disappear in the meantime. Thanks to Vlad Roubtsov for reporting
903 * Make installed scripts executable. Thanks to Jirka Hladky for reporting
905 * Fix libtool revision management when building for Windows. This fix was
906 also released as hwloc v1.4.1.1 Windows builds. Thanks to Hartmut Kaiser
907 for reporting the problem.
908 * Fix the __hwloc_inline keyword in public headers when compiling with a
910 * Add Port info attribute to network OS devices inside OpenFabrics PCI
911 devices so as to identify which interface corresponds to which port.
912 * Document requirements for interoperability helpers: I/O devices discovery
913 is required for some of them; the topology must match the current host
919 * This release contains all changes from v1.3.2.
920 * Fix hwloc_alloc_membind, thanks Karl Napf for reporting the issue.
921 * Fix memory leaks in some get_membind() functions.
922 * Fix helpers converting from Linux libnuma to hwloc (hwloc/linux-libnuma.h)
923 in case of out-of-order NUMA node ids.
924 * Fix some overzealous assertions in the distance grouping code.
925 * Workaround BIOS reporting empty I/O locality in CUDA and OpenFabrics
926 helpers on Linux. Thanks to Albert Solernou for reporting the problem.
927 * Install a valgrind suppressions file hwloc-valgrind.supp (see the FAQ).
928 * Fix memory binding documentation. Thanks to Karl Napf for reporting the
932 Version 1.4.0 (does not contain all v1.3.2 changes)
935 + Add "custom" interface and "assembler" tools to build multi-node
936 topology. See the Multi-node Topologies section in the documentation
938 * Interface improvements
939 + Add symmetric_subtree object attribute to ease assumptions when consulting
940 regular symmetric topologies.
941 + Add a CPUModel and CPUType info attribute to Socket objects on Linux
943 + Add hwloc_get_obj_index_inside_cpuset() to retrieve the "logical" index
944 of an object within a subtree of the topology.
945 + Add more NVIDIA CUDA helpers in cuda.h and cudart.h to find hwloc objects
946 corresponding to CUDA devices.
947 * Discovery improvements
948 + Add a group object above partial distance matrices to make sure
949 the matrices are available in the final topology, except when this
950 new object would contradict the existing hierarchy.
951 + Grouping by distances now also works when loading from XML.
952 + Fix some corner cases in object insertion, for instance when dealing
953 with NUMA nodes without any CPU.
955 + Implement hwloc_get_area_membind() on Linux.
956 + Honor I/O topology flags when importing from XML.
957 + Further improve XML-related error checking and reporting.
958 + Hide synthetic topology error messages unless HWLOC_SYNTHETIC_VERBOSE=1.
960 + Add synthetic exporting of symmetric topologies to lstopo.
961 + lstopo --horiz and --vert can now be applied to some specific object types.
962 + lstopo -v -p now displays distance matrices with physical indexes.
963 + Add hwloc-distances utility to list distances.
965 + Fix and/or document the behavior of most inline functions in hwloc/helper.h
966 when the topology contains some I/O or Misc objects.
967 + Backend documentation enhancements.
969 + Fix missing last bit in hwloc_linux_get_thread_cpubind().
970 Thanks to Carolina Gómez-Tostón Gutiérrez for reporting the issue.
971 + Fix FreeBSD build without cpuid support.
972 + Fix several Windows build issues.
973 + Fix inline keyword definition in public headers.
974 + Fix dependencies in the embedded library.
975 + Improve visibility support detection. Thanks to Dave Love for providing
977 + Remove references to internal symbols in the tools.
982 * This release is only meant to fix the pciutils license issue when upgrading
983 to hwloc v1.4 or later is not possible. It contains several other minor
984 fixes but ignores many of them that are only in v1.4 or later.
985 * Use libpciaccess instead of pciutils/libpci by default for I/O discovery.
986 pciutils/libpci is only used if --enable-libpci is given to configure
987 because its GPL license may taint hwloc. See the Installation section
988 in the documentation for details.
993 * Fix missing last bit in hwloc_linux_get_thread_cpubind().
994 Thanks to Carolina Gómez-Tostón Gutiérrez for reporting the issue.
995 * Fix build with -mcmodel=medium. Thanks to Devendar Bureddy for reporting
997 * Fix build with Solaris Studio 12 compiler when XML is disabled.
998 Thanks to Paul H. Hargrove for reporting the problem.
999 * Fix installation with old GNU sed, for instance on Red Hat 8.
1000 Thanks to Paul H. Hargrove for reporting the problem.
1001 * Fix PCI locality when Linux cgroups restrict the available CPUs.
1002 * Fix floating point issue when grouping by distance on mips64 architecture.
1003 Thanks to Paul H. Hargrove for reporting the problem.
1004 * Fix conversion from/to Linux libnuma when some NUMA nodes have no memory.
1005 * Fix support for gccfss compilers with broken ffs() support. Thanks to
1006 Paul H. Hargrove for reporting the problem and providing a patch.
1007 * Fix FreeBSD build without cpuid support.
1008 * Fix several Windows build issues.
1009 * Fix inline keyword definition in public headers.
1010 * Fix dependencies in the embedded library.
1011 * Detect when a compiler such as xlc may not report compile errors
1012 properly, causing some configure checks to be wrong. Thanks to
1013 Paul H. Hargrove for reporting the problem and providing a patch.
1014 * Improve visibility support detection. Thanks to Dave Love for providing
1016 * Remove references to internal symbols in the tools.
1017 * Fix installation on systems with limited command-line size.
1018 Thanks to Paul H. Hargrove for reporting the problem.
1019 * Further improve XML-related error checking and reporting.
1024 * Fix pciutils detection with pkg-config when not installed in standard
1026 * Fix visibility options detection with the Solaris Studio compiler.
1027 Thanks to Igor Galić and Terry Dontje for reporting the problems.
1028 * Fix support for old Linux sched.h headers such as those found
1029 on Red Hat 8. Thanks to Paul H. Hargrove for reporting the problems.
1030 * Fix inline and attribute support for Solaris compilers. Thanks to
1031 Dave Love for reporting the problems.
1032 * Print a short summary at the end of the configure output. Thanks to
1033 Stefan Eilemann for the suggestion.
1034 * Add --disable-libnuma configure option to disable libnuma-based
1035 memory binding support on Linux. Thanks to Rayson Ho for the
1037 * Make hwloc's configure script properly obey $PKG_CONFIG. Thanks to
1038 Nathan Phillip Brink for raising the issue.
1039 * Silence some harmless pciutils warnings, thanks to Paul H. Hargrove
1040 for reporting the problem.
1041 * Fix the documentation with respect to hwloc_pid_t and hwloc_thread_t
1042 being either pid_t and pthread_t on Unix, or HANDLE on Windows.
1048 + Add I/O devices and bridges to the topology using the pciutils
1049 library. Only enabled after setting the relevant flag with
1050 hwloc_topology_set_flags() before hwloc_topology_load(). See the
1051 I/O Devices section in the documentation for details.
1052 * Discovery improvements
1053 + Add associativity to the cache attributes.
1054 + Add support for s390/z11 "books" on Linux.
1055 + Add the HWLOC_GROUPING_ACCURACY environment variable to relax
1056 distance-based grouping constraints. See the Environment Variables
1057 section in the documentation for details about grouping behavior
1059 + Allow user-given distance matrices to remove or replace those
1060 discovered by the OS backend.
1062 + XML is now always supported: a minimalistic custom import/export
1063 code is used when libxml2 is not available. It is only guaranteed
1064 to read XML files generated by hwloc.
1065 + hwloc_topology_export_xml() and export_xmlbuffer() now return an
1067 + Add hwloc_free_xmlbuffer() to free the buffer allocated by
1068 hwloc_topology_export_xmlbuffer().
1069 + Hide XML topology error messages unless HWLOC_XML_VERBOSE=1.
1071 + Add hwloc_obj_add_info to customize object info attributes.
1073 + lstopo now displays I/O devices by default. Several options are
1074 added to configure the I/O discovery.
1075 + hwloc-calc and hwloc-bind now accept I/O devices as input.
1076 + Add --restrict option to hwloc-calc and hwloc-distribute.
1077 + Add --sep option to change the output field separator in hwloc-calc.
1078 + Add --whole-system option to hwloc-ps.
1083 * Fix build on AIX 5.2, thanks Utpal Kumar Ray for the report.
1084 * Fix XML import of very large page sizes or counts on 32bits platform,
1085 thanks to Karsten Hopp for the RedHat ticket.
1086 * Fix crash when administrator limitations such as Linux cgroup require
1087 to restrict distance matrices. Thanks to Ake Sandgren for reporting the
1089 * Fix the removal of objects such as AMD Magny-Cours dual-node sockets
1090 in case of administrator restrictions.
1091 * Improve error reporting and messages in case of wrong synthetic topology
1093 * Several other minor internal fixes and documentation improvements.
1098 * Improve support of AMD Bulldozer "Compute-Unit" modules by detecting
1099 logical processors with different core IDs on Linux.
1100 * Fix hwloc-ps crash when listing processes from another Linux cpuset.
1101 Thanks to Carl Smith for reporting the problem.
1102 * Fix build on AIX and Solaris. Thanks to Carl Smith and Andreas Kupries
1103 for reporting the problems.
1104 * Fix cache size detection on Darwin. Thanks to Erkcan Özcan for reporting
1106 * Make configure fail if --enable-xml or --enable-cairo is given and
1107 proper support cannot be found. Thanks to Andreas Kupries for reporting
1109 * Fix spurious L1 cache detection on AIX. Thanks to Hendryk Bockelmann
1110 for reporting the problem.
1111 * Fix hwloc_get_last_cpu_location(THREAD) on Linux. Thanks to Gabriele
1112 Fatigati for reporting the problem.
1113 * Fix object distance detection on Solaris.
1114 * Add pthread_self weak symbol to ease static linking.
1115 * Minor documentation fixes.
1121 + Expose latency matrices in the API as an array of distance structures
1122 within objects. Add several helpers to find distances.
1123 + Add hwloc_topology_set_distance_matrix() and environment variables
1124 to provide a matrix of distances between a given set of objects.
1125 + Add hwloc_get_last_cpu_location() and hwloc_get_proc_last_cpu_location()
1126 to retrieve the processors where a process or thread recently ran.
1127 - Add the corresponding --get-last-cpu-location option to hwloc-bind.
1128 + Add hwloc_topology_restrict() to restrict an existing topology to a
1130 - Add the corresponding --restrict option to lstopo.
1132 + Add hwloc_bitmap_list_sscanf/snprintf/asprintf to convert between bitmaps
1133 and strings such as 4-5,7-9,12,15-
1134 + hwloc_bitmap_set/clr_range() now support infinite ranges.
1135 + Clarify the difference between inserting Misc objects by cpuset or by
1137 + hwloc_insert_misc_object_by_cpuset() now returns NULL in case of error.
1138 * Discovery improvements
1139 + x86 backend (for freebsd): add x2APIC support
1140 + Support standard device-tree phandle, to get better support on e.g. ARM
1141 systems providing it.
1142 + Detect cache size on AIX. Thanks Christopher and IBM.
1143 + Improve grouping to support asymmetric topologies.
1145 + Command-line tools now support "all" and "root" special locations
1146 consisting in the entire topology, as well as type names with depth
1147 attributes such as L2 or Group4.
1148 + hwloc-calc improvements:
1149 - Add --number-of/-N option to report the number of objects of a given
1151 - -I is now equivalent to --intersect for listing the indexes of
1152 objects of a given type or depth that intersects the input.
1153 - Add -H to report the output as a hierarchical combination of types
1155 + Add --thissystem to lstopo.
1156 + Add lstopo-win, a console-less lstopo variant on Windows.
1158 + Remove C99 usage from code base.
1159 + Rename hwloc-gather-topology.sh into hwloc-gather-topology
1160 + Fix AMD cache discovery on freebsd when there is no L3 cache, thanks
1161 Andriy Gapon for the fix.
1166 * Fix a segfault in the distance-based grouping code when some objects
1167 are not placed in any group. Thanks to Bernd Kallies for reporting
1168 the problem and providing a patch.
1169 * Fix the command-line parsing of hwloc-bind --mempolicy interleave.
1170 Thanks to Guy Streeter for reporting the problem.
1171 * Stop truncating the output in hwloc_obj_attr_snprintf() and in the
1172 corresponding lstopo output. Thanks to Guy Streeter for reporting the
1174 * Fix object levels ordering in synthetic topologies.
1175 * Fix potential incoherency between device tree and kernel information,
1176 when SMT is disabled on Power machines.
1177 * Fix and document the behavior of hwloc_topology_set_synthetic() in case
1178 of invalid argument. Thanks to Guy Streeter for reporting the problem.
1179 * Add some verbose error message reporting when it looks like the OS
1180 gives erroneous information.
1181 * Do not include unistd.h and stdint.h in public headers on Windows.
1182 * Move config.h files into their own subdirectories to avoid name
1183 conflicts when AC_CONFIG_HEADERS adds -I's for them.
1184 * Remove the use of declaring variables inside "for" loops.
1185 * Some other minor fixes.
1186 * Many minor documentation fixes.
1191 * Add hwloc_get_api_version() which returns the version of hwloc used
1192 at runtime. Thanks to Guy Streeter for the suggestion.
1193 * Fix the number of hugepages reported for NUMA nodes on Linux.
1194 * Fix hwloc_bitmap_to_ulong() right after allocating the bitmap.
1195 Thanks to Bernd Kallies for reporting the problem.
1196 * Fix hwloc_bitmap_from_ith_ulong() to properly zero the first ulong.
1197 Thanks to Guy Streeter for reporting the problem.
1198 * Fix hwloc_get_membind_nodeset() on Linux.
1199 Thanks to Bernd Kallies for reporting the problem and providing a patch.
1200 * Fix some file descriptor leaks in the Linux discovery.
1201 * Fix the minimum width of NUMA nodes, caches and the legend in the graphical
1202 lstopo output. Thanks to Jirka Hladky for reporting the problem.
1203 * Various fixes to bitmap conversion from/to taskset-strings.
1204 * Fix and document snprintf functions behavior when the buffer size is too
1205 small or zero. Thanks to Guy Streeter for reporting the problem.
1206 * Fix configure to avoid spurious enabling of the cpuid backend.
1207 Thanks to Tim Anderson for reporting the problem.
1208 * Cleanup error management in hwloc-gather-topology.sh.
1209 Thanks to Jirka Hladky for reporting the problem and providing a patch.
1210 * Add a manpage and usage for hwloc-gather-topology.sh on Linux.
1211 Thanks to Jirka Hladky for providing a patch.
1212 * Memory binding documentation enhancements.
1219 + Increase HWLOC_API_VERSION to 0x00010100 so that API changes may be
1220 detected at build-time.
1221 + Add a memory binding interface.
1222 + The cpuset API (hwloc/cpuset.h) is now deprecated. It is replaced by
1223 the bitmap API (hwloc/bitmap.h) which offers the same features with more
1224 generic names since it applies to CPU sets, node sets and more.
1225 Backward compatibility with the cpuset API and ABI is still provided but
1226 it will be removed in a future release.
1227 Old types (hwloc_cpuset_t, ...) are still available as a way to clarify
1228 what kind of hwloc_bitmap_t each API function manipulates.
1229 Upgrading to the new API only requires to replace hwloc_cpuset_ function
1230 calls with the corresponding hwloc_bitmap_ calls, with the following
1231 renaming exceptions:
1232 - hwloc_cpuset_cpu -> hwloc_bitmap_only
1233 - hwloc_cpuset_all_but_cpu -> hwloc_bitmap_allbut
1234 - hwloc_cpuset_from_string -> hwloc_bitmap_sscanf
1235 + Add an `infos' array in each object to store couples of info names and
1236 values. It enables generic storage of things like the old dmi board infos
1237 that were previously stored in machine specific attributes.
1238 + Add linesize cache attribute.
1240 + Bitmaps (and thus CPU sets and node sets) are dynamically (re-)allocated,
1241 the maximal number of CPUs (HWLOC_NBMAXCPUS) has been removed.
1242 + Improve the distance-based grouping code to better support irregular
1244 + Add support for device-tree to get cache information (useful on Power
1247 + Add NVIDIA CUDA helpers in cuda.h and cudart.h to ease interoperability
1248 with CUDA Runtime and Driver APIs.
1249 + Add Myrinet Express helper in myriexpress.h to ease interoperability.
1251 + lstopo now displays physical/OS indexes by default in graphical mode
1252 (use -l to switch back to logical indexes). The textual output still uses
1253 logical by default (use -p to switch to physical indexes).
1254 + lstopo prefixes logical indexes with `L#' and physical indexes with `P#'.
1255 Physical indexes are also printed as `P#N' instead of `phys=N' within
1256 object attributes (in parentheses).
1257 + Add a legend at the bottom of the lstopo graphical output, use --no-legend
1259 + Add hwloc-ps to list process' bindings.
1260 + Add --membind and --mempolicy options to hwloc-bind.
1261 + Improve tools command-line options by adding a generic --input option
1262 (and more) which replaces the old --xml, --synthetic and --fsys-root.
1263 + Cleanup lstopo output configuration by adding --output-format.
1264 + Add --intersect in hwloc-calc, and replace --objects with --largest.
1265 + Add the ability to work on standard input in hwloc-calc.
1266 + Add --from, --to and --at in hwloc-distrib.
1267 + Add taskset-specific functions and command-line tools options to
1268 manipulate CPU set strings in the format of the taskset program.
1269 + Install hwloc-gather-topology.sh on Linux.
1275 * Fix support for Linux cpuset when emulated by a cgroup mount point.
1276 * Remove unneeded runtime dependency on libibverbs.so in the library and
1278 * Fix hwloc_cpuset_to_linux_libnuma_ulongs in case of non-linear OS-indexes
1280 * lstopo now displays physical/OS indexes by default in graphical mode
1281 (use -l to switch back to logical indexes). The textual output still uses
1282 logical by default (use -p to switch to physical indexes).
1288 * Public headers can now be included directly from C++ programs.
1289 * Solaris fix for non-contiguous cpu numbers. Thanks to Rolf vandeVaart for
1290 reporting the issue.
1291 * Darwin 10.4 fix. Thanks to Olivier Cessenat for reporting the issue.
1292 * Revert 1.0.1 patch that ignored sockets with unknown ID values since it
1293 only slightly helped POWER7 machines with old Linux kernels while it
1294 prevents recent kernels from getting the complete POWER7 topology.
1295 * Fix hwloc_get_common_ancestor_obj().
1296 * Remove arch-specific bits in public headers.
1297 * Some fixes in the lstopo graphical output.
1298 * Various man page clarifications and minor updates.
1304 * Various Solaris fixes. Thanks to Yannick Martin for reporting the issue.
1305 * Fix "non-native" builds on x86 platforms (e.g., when building 32
1306 bit executables with compilers that natively build 64 bit).
1307 * Ignore sockets with unknown ID values (which fixes issues on POWER7
1308 machines). Thanks to Greg Bauer for reporting the issue.
1309 * Various man page clarifications and minor updates.
1310 * Fixed memory leaks in hwloc_setup_group_from_min_distance_clique().
1311 * Fix cache type filtering on MS Windows 7. Thanks to Αλέξανδρος
1312 Παπαδογιαννάκ for reporting the issue.
1313 * Fixed warnings when compiling with -DNDEBUG.
1319 * The ABI of the library has changed.
1321 + Add FreeBSD support.
1322 + Add x86 cpuid based backend.
1323 + Add Linux cgroup support to the Linux cpuset code.
1324 + Support binding of entire multithreaded process on Linux.
1325 + Fix and enable Group support in Windows.
1326 + Cleanup XML export/import.
1328 + HWLOC_OBJ_PROC is renamed into HWLOC_OBJ_PU for "Processing Unit",
1329 its stringified type name is now "PU".
1330 + Use new HWLOC_OBJ_GROUP objects instead of MISC when grouping
1331 objects according to NUMA distances or arbitrary OS aggregation.
1332 + Rework memory attributes.
1333 + Add different cpusets in each object to specify processors that
1334 are offline, unavailable, ...
1335 + Cleanup the storage of object names and DMI infos.
1337 + Add support for looking up specific PID topology information.
1338 + Add hwloc_topology_export_xml() to export the topology in a XML file.
1339 + Add hwloc_topology_get_support() to retrieve the supported features
1340 for the current topology context.
1341 + Support non-SYSTEM object as the root of the tree, use MACHINE in
1343 + Add hwloc_get_*cpubind() routines to retrieve the current binding
1344 of processes and threads.
1346 + Add HWLOC_API_VERSION to help detect the currently used API version.
1347 + Add missing ending "e" to *compare* functions.
1348 + Add several routines to emulate PLPA functions.
1349 + Rename and rework the cpuset and/or/xor/not/clear operators to output
1350 their result in a dedicated argument instead of modifying one input.
1351 + Deprecate hwloc_obj_snprintf() in favor of hwloc_obj_type/attr_snprintf().
1352 + Clarify the use of parent and ancestor in the API, do not use father.
1353 + Replace hwloc_get_system_obj() with hwloc_get_root_obj().
1354 + Return -1 instead of HWLOC_OBJ_TYPE_MAX in the API since the latter
1356 + Relax constraints in hwloc_obj_type_of_string().
1357 + Improve displaying of memory sizes.
1358 + Add 0x prefix to cpuset strings.
1360 + lstopo now displays logical indexes by default, use --physical to
1361 revert back to OS/physical indexes.
1362 + Add colors in the lstopo graphical outputs to distinguish between online,
1363 offline, reserved, ... objects.
1364 + Extend lstopo to show cpusets, filter objects by type, ...
1365 + Renamed hwloc-mask into hwloc-calc which supports many new options.
1367 + Add a hwloc(7) manpage containing general information.
1368 + Add documentation about how to switch from PLPA to hwloc.
1369 + Cleanup the distributed documentation files.
1371 + Many compilers warning fixes.
1372 + Cleanup the ABI by using the visibility attribute.
1373 + Add project embedding support.
1376 Version 0.9.4 (unreleased)
1377 --------------------------
1379 * Fix reseting colors to normal in lstopo -.txt output.
1380 * Fix Linux pthread_t binding error report.
1386 * Fix autogen.sh to work with Autoconf 2.63.
1387 * Fix various crashes in particular conditions:
1388 - xml files with root attributes
1390 - partial sysfs support
1391 - unparseable /proc/cpuinfo
1392 - ignoring NUMA level while Misc level have been generated
1393 * Tweak documentation a bit
1394 * Do not require the pthread library for binding the current thread on Linux
1395 * Do not erroneously consider the sched_setaffinity prototype is the old version
1396 when there is actually none.
1397 * Fix _syscall3 compilation on archs for which we do not have the
1398 sched_setaffinity system call number.
1400 * Fix libraries dependencies: now only lstopo depends on libtermcap, fix
1402 * Have make check always build and run hwloc-hello.c
1403 * Do not limit size of a cpuset.
1409 * Trivial documentation changes.
1415 * Re-branded to "hwloc" and moved to the Open MPI project, relicensed under the
1417 * The prefix of all functions and tools is now hwloc, and some public
1418 functions were also renamed for real.
1419 * Group NUMA nodes into Misc objects according to their physical distance
1420 that may be reported by the OS/BIOS.
1421 May be ignored by setting HWLOC_IGNORE_DISTANCES=1 in the environment.
1422 * Ignore offline CPUs on Solaris.
1423 * Improved binding support on AIX.
1424 * Add HP-UX support.
1425 * CPU sets are now allocated/freed dynamically.
1426 * Add command line options to tune the lstopo graphical output, add
1427 semi-graphical textual output
1428 * Extend topobind to support multiple cpusets or objects on the command
1429 line as topomask does.
1430 * Add an Infiniband-specific helper hwloc/openfabrics-verbs.h to retrieve
1431 the physical location of IB devices.
1434 Version 0.9 (libtopology)
1435 -------------------------