1Copyright © 2009 CNRS 2Copyright © 2009-2017 Inria. All rights reserved. 3Copyright © 2009-2013 Université Bordeaux 4Copyright © 2009-2011 Cisco Systems, Inc. All rights reserved. 5 6$COPYRIGHT$ 7 8Additional copyrights may follow 9 10$HEADER$ 11 12=========================================================================== 13 14This file contains the main features as well as overviews of specific 15bug fixes (and other actions) for each version of hwloc since version 160.9 (as initially released as "libtopology", then re-branded to "hwloc" 17in v0.9.1). 18 19 20Version 1.11.7 21-------------- 22* Fix hwloc-bind --membind for CPU-less NUMA nodes (again). 23 Thanks to Gilles Gouaillardet for reporting the issue. 24* Fix a memory leak on IBM S/390 platforms running Linux. 25* Fix a memory leak when forcing the x86 backend first on amd64/topoext 26 platforms running Linux. 27* Command-line tools now support "hbm" instead "numanode" for filtering 28 only high-bandwidth memory nodes when selecting locations. 29 + hwloc-bind also support --hbm and --no-hbm for filtering only or 30 no HBM nodes. 31 Thanks to Nicolas Denoyelle for the suggestion. 32* Add --children and --descendants to hwloc-info for listing object 33 children or object descendants of a specific type. 34* Add --no-index, --index, --no-attrs, --attrs to disable/enable display 35 of index numbers or attributes in the graphical lstopo output. 36* Try to gather hwloc-dump-hwdata output from all possible locations 37 in hwloc-gather-topology. 38* Updates to the documentation of locations in hwloc(7) and 39 command-line tools manpages. 40 41 42Version 1.11.6 43-------------- 44* Make the Linux discovery about twice faster, especially on the CPU side, 45 by trying to avoid sysfs file accesses as much as possible. 46* Add support for AMD Family 17h processors (Zen) SMT cores in the Linux 47 and x86 backends. 48* Add the HWLOC_TOPOLOGY_FLAG_THISSYSTEM_ALLOWED_RESOURCES flag (and the 49 HWLOC_THISSYSTEM_ALLOWED_RESOURCES environment variable) for reading the 50 set of allowed resources from the local operating system even if the 51 topology was loaded from XML or synthetic. 52* Fix hwloc_bitmap_set/clr_range() for infinite ranges that do not 53 overlap currently defined ranges in the bitmap. 54* Don't reset the lstopo zoom scale when moving the X11 window. 55* lstopo now has --flags for manually setting topology flags. 56* hwloc_get_depth_type() returns HWLOC_TYPE_DEPTH_UNKNOWN for Misc objects. 57 58 59Version 1.11.5 60-------------- 61* Add support for Knights Mill Xeon Phi, thanks to Piotr Luc for the patch. 62* Reenable distance gathering on Solaris, disabled by mistake since v1.0. 63 Thanks to TU Wien for the help. 64* Fix hwloc_get_*obj*_inside_cpuset() functions to ignore objects with 65 empty CPU sets, for instance, CPU-less NUMA nodes such as KNL MCDRAM. 66 Thanks to Nicolas Denoyelle for the report. 67* Fix XML import of multiple distance matrices. 68* Add a FAQ entry about "hwloc is only a structural model, it ignores 69 performance models, memory bandwidth, etc.?" 70 71 72Version 1.11.4 73-------------- 74* Add MemoryMode and ClusterMode attributes in the Machine object on KNL. 75 Add doc/examples/get-knl-modes.c for an example of retrieving them. 76 Thanks to Grzegorz Andrejczuk. 77* Fix Linux build with -m32 with respect to libudev. 78 Thanks to Paul Hargrove for reporting the issue. 79* Fix build with Visual Studio 2015, thanks to Eloi Gaudry for reporting 80 the issue and providing the patch. 81* Don't forget to display OS device children in the graphical lstopo. 82* Fix a memory leak on Solaris, thanks to Bryon Gloden for the patch. 83* Properly handle realloc() failures, thanks to Bryon Gloden for reporting 84 the issue. 85* Fix lstopo crash in ascii/fig/windows outputs when some objects have a 86 lstopoStyle info attribute. 87 88 89Version 1.11.3 90-------------- 91* Bug fixes 92 + Fix a memory leak on Linux S/390 hosts with books. 93 + Fix /proc/mounts parsing on Linux by using mntent.h. 94 Thanks to Nathan Hjelm for reporting the issue. 95 + Fix a x86 infinite loop on VMware due to the x2APIC feature being 96 advertised without actually being fully supported. 97 Thanks to Jianjun Wen for reporting the problem and testing the patch. 98 + Fix the return value of hwloc_alloc() on mmap() failure. 99 Thanks to Hugo Brunie for reporting the issue. 100 + Fix the return value of command-line tools in some error cases. 101 + Do not break individual thread bindings during x86 backend discovery in a 102 multithreaded process. Thanks to Farouk Mansouri for the report. 103 + Fix hwloc-bind --membind for CPU-less NUMA nodes. 104 + Fix some corner cases in the XML export/import of application userdata. 105* API Improvements 106 + Add HWLOC_MEMBIND_BYNODESET flag so that membind() functions accept 107 either cpusets or nodesets. 108 + Add hwloc_get_area_memlocation() to check where pages are actually 109 allocated. Only implemented on Linux for now. 110 - There's no _nodeset() variant, but the new flag HWLOC_MEMBIND_BYNODESET 111 is supported. 112 + Make hwloc_obj_type_sscanf() parse back everything that may be outputted 113 by hwloc_obj_type_snprintf(). 114* Detection Improvements 115 + Allow the x86 backend to add missing cache levels, so that it completes 116 what the Solaris backend lacks. 117 Thanks to Ryan Zezeski for reporting the issue. 118 + Do not filter-out FibreChannel PCI adapters by default anymore. 119 Thanks to Matt Muggeridge for the report. 120 + Add support for CUDA compute capability 6.x. 121* Tools 122 + Add --support to hwloc-info to list supported features, just like with 123 hwloc_topology_get_support(). 124 - Also add --objects and --topology to explicitly switch between the 125 default modes. 126 + Add --tid to let hwloc-bind operate on individual threads on Linux. 127 + Add --nodeset to let hwloc-bind report memory binding as NUMA node sets. 128 + hwloc-annotate and lstopo don't drop application userdata from XMLs anymore. 129 - Add --cu to hwloc-annotate to drop these application userdata. 130 + Make the hwloc-dump-hwdata dump directory configurable through configure 131 options such as --runstatedir or --localstatedir. 132* Misc Improvements 133 + Add systemd service template contrib/systemd/hwloc-dump-hwdata.service 134 for launching hwloc-dump-hwdata at boot on Linux. 135 Thanks to Grzegorz Andrejczuk. 136 + Add HWLOC_PLUGINS_BLACKLIST environment variable to prevent some plugins 137 from being loaded. Thanks to Alexandre Denis for the suggestion. 138 + Small improvements for various Windows build systems, 139 thanks to Jonathan L Peyton and Marco Atzeri. 140 141 142Version 1.11.2 143-------------- 144* Improve support for Intel Knights Landing Xeon Phi on Linux: 145 + Group local NUMA nodes of normal memory (DDR) and high-bandwidth memory 146 (MCDRAM) together through "Cluster" groups so that the local MCDRAM is 147 easy to find. 148 - See "How do I find the local MCDRAM NUMA node on Intel Knights 149 Landing Xeon Phi?" in the documentation. 150 - For uniformity across all KNL configurations, always have a NUMA node 151 object even if the host is UMA. 152 + Fix the detection of the memory-side cache: 153 - Add the hwloc-dump-hwdata superuser utility to dump SMBIOS information 154 into /var/run/hwloc/ as root during boot, and load this dumped 155 information from the hwloc library at runtime. 156 - See "Why do I need hwloc-dump-hwdata for caches on Intel Knights 157 Landing Xeon Phi?" in the documentation. 158 Thanks to Grzegorz Andrejczuk for the patches and for the help. 159* The x86 and linux backends may now be combined for discovering CPUs 160 through x86 CPUID and memory from the Linux kernel. 161 This is useful for working around buggy CPU information reported by Linux 162 (for instance the AMD Bulldozer/Piledriver bug below). 163 Combination is enabled by passing HWLOC_COMPONENTS=x86 in the environment. 164* Fix L3 cache sharing on AMD Opteron 63xx (Piledriver) and 62xx (Bulldozer) 165 in the x86 backend. Thanks to many users who helped. 166* Fix the overzealous L3 cache sharing fix added to the x86 backend in 1.11.1 167 for AMD Opteron 61xx (Magny-Cours) processors. 168* The x86 backend may now add the info attribute Inclusive=0 or 1 to caches 169 it discovers, or to caches discovered by other backends earlier. 170 Thanks to Guillaume Beauchamp for the patch. 171* Fix the management on alloc_membind() allocation failures on AIX, HP-UX 172 and OSF/Tru64. 173* Fix spurious failures to load with ENOMEM on AIX in case of Misc objects 174 below PUs. 175* lstopo improvements in X11 and Windows graphical mode: 176 + Add + - f 1 shortcuts to manually zoom-in, zoom-out, reset the scale, 177 or fit the entire window. 178 + Display all keyboard shortcuts in the console. 179* Debug messages may be disabled at runtime by passing HWLOC_DEBUG_VERBOSE=0 180 in the environment when --enable-debug was passed to configure. 181* Add a FAQ entry "What are these Group objects in my topology?". 182 183 184Version 1.11.1 185-------------- 186* Detection fixes 187 + Hardwire the topology of Fujitsu K-computer, FX10, FX100 servers to 188 workaround buggy Linux kernels. 189 Thanks to Takahiro Kawashima and Gilles Gouaillardet. 190 + Fix L3 cache information on AMD Opteron 61xx Magny-Cours processors 191 in the x86 backend. Thanks to Guillaume Beauchamp for the patch. 192 + Detect block devices directly attached to PCI without a controller, 193 for instance NVMe disks. Thanks to Barry M. Tannenbaum. 194 + Add the PCISlot attribute to all PCI functions instead of only the 195 first one. 196* Miscellaneous internal fixes 197 + Ignore PCI bridges that could fail assertions by reporting buggy 198 secondary-subordinate bus numbers 199 Thanks to George Bosilca for reporting the issue. 200 + Fix an overzealous assertion when inserting an intermediate Group object 201 while Groups are totally ignored. 202 + Fix a memory leak on Linux on AMD processors with dual-core compute units. 203 Thanks to Bob Benner. 204 + Fix a memory leak on failure to load a xml diff file. 205 + Fix some segfaults when inputting an invalid synthetic description. 206 + Fix a segfault when plugins fail to find core symbols. 207 Thanks to Guy Streeter. 208* Many fixes and improvements in the Windows backend: 209 + Fix the discovery of more than 32 processors and multiple processor 210 groups. Thanks to Barry M. Tannenbaum for the help. 211 + Add thread binding set support in case of multiple process groups. 212 + Add thread binding get support. 213 + Add get_last_cpu_location() support for the current thread. 214 + Disable the unsupported process binding in case of multiple processor 215 groups. 216 + Fix/update the Visual Studio support under contrib/windows. 217 Thanks to Eloi Gaudry for the help. 218* Tools fixes 219 + Fix a segfault when displaying logical indexes in the graphical lstopo. 220 Thanks to Guillaume Mercier for reporting the issue. 221 + Fix lstopo linking with X11 libraries, for instance on Mac OS X. 222 Thanks to Scott Atchley and Pierre Ramet for reporting the issue. 223 + hwloc-annotate, hwloc-diff and hwloc-patch do not drop unavailable 224 resources from the output anymore and those may be annotated as well. 225 + Command-line tools may now import XML from the standard input with -i -.xml 226 + Add missing documentation for the hwloc-info --no-icaches option. 227 228 229Version 1.11.0 230-------------- 231* API 232 + Socket objects are renamed into Package to align with the terminology 233 used by processor vendors. The old HWLOC_OBJ_SOCKET type and "Socket" 234 name are still supported for backward compatibility. 235 + HWLOC_OBJ_NODE is replaced with HWLOC_OBJ_NUMANODE for clarification. 236 HWLOC_OBJ_NODE is still supported for backward compatibility. 237 "Node" and "NUMANode" strings are supported as in earlier releases. 238* Detection improvements 239 + Add support for Intel Knights Landing Xeon Phi. 240 Thanks to Grzegorz Andrejczuk and Lukasz Anaczkowski. 241 + Add Vendor, Model, Revision, SerialNumber, Type and LinuxDeviceID 242 info attributes to Block OS devices on Linux. Thanks to Vineet Pedaballe 243 for the help. 244 - Add --disable-libudev to avoid dependency on the libudev library. 245 + Add "MemoryModule" Misc objects with information about DIMMs, on Linux 246 when privileged and when I/O is enabled. 247 Thanks to Vineet Pedaballe for the help. 248 + Add a PCISlot attribute to PCI devices on Linux when supported to 249 identify the physical PCI slot where the board is plugged. 250 + Add CPUStepping info attribute on x86 processors, 251 thanks to Thomas Röhl for the suggestion. 252 + Ignore the device-tree on non-Power architectures to avoid buggy 253 detection on ARM. Thanks to Orion Poplawski for reporting the issue. 254 + Work-around buggy Xeon E5v3 BIOS reporting invalid PCI-NUMA affinity 255 for the PCI links on the second processor. 256 + Add support for CUDA compute capability 5.x, thanks Benjamin Worpitz. 257 + Many fixes to the x86 backend 258 - Add L1i and fix L2/L3 type on old AMD processors without topoext support. 259 - Fix Intel CPU family and model numbers when basic family isn't 6 or 15. 260 - Fix package IDs on recent AMD processors. 261 - Fix misc issues due to incomplete APIC IDs on x2APIC processors. 262 - Avoid buggy discovery on old SGI Altix UVs with non-unique APIC IDs. 263 + Gather total machine memory on NetBSD. 264* Tools 265 + lstopo 266 - Collapse identical PCI devices unless --no-collapse is given. 267 This avoids gigantic outputs when a PCI device contains dozens of 268 identical virtual functions. 269 - The ASCII art output is now called "ascii", for instance in 270 "lstopo -.ascii". 271 The former "txt" extension is retained for backward compatibility. 272 - Automatically scales graphical box width to the inner text in Cairo, 273 ASCII and Windows outputs. 274 - Add --rect to lstopo to force rectangular layout even for NUMA nodes. 275 - Add --restrict-flags to configure the behavior of --restrict. 276 - Objects may have a "Type" info attribute to specify a better type name 277 and display it in lstopo. 278 - Really export all verbose information to the given output file. 279 + hwloc-annotate 280 - May now operate on all types of objects, including I/O. 281 - May now insert Misc objects in the topology. 282 - Do not drop instruction caches and I/O devices from the output anymore. 283 + Fix lstopo path in hwloc-gather-topology after install. 284* Misc 285 + Fix hwloc/cudart.h for machines with multiple PCI domains, 286 thanks to Imre Kerr for reporting the problem. 287 + Fix PCI Bridge-specific depth attribute. 288 + Fix hwloc_bitmap_intersect() for two infinite bitmaps. 289 + Fix some corner cases in the building of levels on large NUMA machines 290 with non-uniform NUMA groups and I/Os. 291 + Improve the performance of object insertion by cpuset for large 292 topologies. 293 + Prefix verbose XML import errors with the source name. 294 + Improve pkg-config checks and error messages. 295 + Fix excluding after a component with an argument in the HWLOC_COMPONENTS 296 environment variable. 297* Documentation 298 + Fix the recommended way in documentation and examples to allocate memory 299 on some node, it should use HWLOC_MEMBIND_BIND. 300 Thanks to Nicolas Bouzat for reporting the issue. 301 + Add a "Miscellaneous objects" section in the documentation. 302 + Add a FAQ entry "What happens to my topology if I disable symmetric 303 multithreading, hyper-threading, etc. ?" to the documentation. 304 305 306Version 1.10.1 307-------------- 308* Actually remove disallowed NUMA nodes from nodesets when the whole-system 309 flag isn't enabled. 310* Fix the gathering of PCI domains. Thanks to James Custer for reporting 311 the issue and providing a patch. 312* Fix the merging of identical parent and child in presence of Misc objects. 313 Thanks to Dave Love for reporting the issue. 314* Fix some misordering of children when merging with ignore_keep_structure() 315 in partially allowed topologies. 316* Fix an overzealous assertion in the debug code when running on a single-PU 317 host with I/O. Thanks to Thomas Van Doren for reporting the issue. 318* Don't forget to setup NUMA node object nodesets in x86 backend (for BSDs) 319 and OSF/Tru64 backend. 320* Fix cpuid-x86 build error with gcc -O3 on x86-32. Thanks to Thomas Van Doren 321 for reporting the issue. 322* Fix support for future very large caches in the x86 backend. 323* Fix vendor/device names for SR-IOV PCI devices on Linux. 324* Fix an unlikely crash in case of buggy hierarchical distance matrix. 325* Fix PU os_index on some AIX releases. Thanks to Hendryk Bockelmann and 326 Erik Schnetter for helping debugging. 327* Fix hwloc_bitmap_isincluded() in case of infinite sets. 328* Change hwloc-ls.desktop into a lstopo.desktop and only install it if 329 lstopo is built with Cairo/X11 support. It cannot work with a non-graphical 330 lstopo or hwloc-ls. 331* Add support for the renaming of Socket into Package in future releases. 332* Add support for the replacement of HWLOC_OBJ_NODE with HWLOC_OBJ_NUMANODE 333 in future releases. 334* Clarify the documentation of distance matrices in hwloc.h and in the manpage 335 of the hwloc-distances. Thanks to Dave Love for the suggestion. 336* Improve some error messages by displaying more information about the 337 hwloc library in use. 338* Document how to deal with the ABI break when upgrading to the upcoming 2.0 339 See "How do I handle ABI breaks and API upgrades ?" in the FAQ. 340 341 342Version 1.10.0 343-------------- 344* API 345 + Add hwloc_topology_export_synthetic() to export a topology to a 346 synthetic string without using lstopo. See the Synthetic topologies 347 section in the documentation. 348 + Add hwloc_topology_set/get_userdata() to let the application save 349 a private pointer in the topology whenever it needs a way to find 350 its own object corresponding to a topology. 351 + Add hwloc_get_numanode_obj_by_os_index() and document that this function 352 as well as hwloc_get_pu_obj_by_os_index() are good at converting 353 nodesets and cpusets into objects. 354 + hwloc_distrib() does not ignore any objects anymore when there are 355 too many of them. They get merged with others instead. 356 Thanks to Tim Creech for reporting the issue. 357* Tools 358 + hwloc-bind --get <command-line> now executes the command after displaying 359 the binding instead of ignoring the command entirely. 360 Thanks to John Donners for the suggestion. 361 + Clarify that memory sizes shown in lstopo are local by default 362 unless specified (total memory added in the root object). 363* Synthetic topologies 364 + Synthetic topology descriptions may now specify attributes such as 365 memory sizes and OS indexes. See the Synthetic topologies section 366 in the documentation. 367 + lstopo now exports in this fully-detailed format by default. 368 The new option --export-synthetic-flags may be used to revert 369 back the old format. 370* Documentation 371 + Add the doc/examples/ subdirectory with several real-life examples, 372 including the already existing hwloc-hello.C for basics. 373 Thanks to Rob Aulwes for the suggestion. 374 + Improve the documentation of CPU and memory binding in the API. 375 + Add a FAQ entry about operating system errors, especially on AMD 376 platforms with buggy cache information. 377 + Add a FAQ entry about loading many topologies in a single program. 378* Misc 379 + Work around buggy Linux kernels reporting 2 sockets instead 380 1 socket with 2 NUMA nodes for each Xeon E5 v3 (Haswell) processor. 381 + pciutils/libpci support is now removed since libpciaccess works 382 well and there's also a Linux-specific PCI backend. For the record, 383 pciutils was GPL and therefore disabled by default since v1.6.2. 384 + Add --disable-cpuid configure flag to work around buggy processor 385 simulators reporting invalid CPUID information. 386 Thanks for Andrew Friedley for reporting the issue. 387 + Fix a racy use of libltdl when manipulating multiple topologies in 388 different threads. 389 Thanks to Andra Hugo for reporting the issue and testing patches. 390 + Fix some build failures in private/misc.h. 391 Thanks to Pavan Balaji and Ralph Castain for the reports. 392 + Fix failures to detect X11/Xutil.h on some Solaris platforms. 393 Thanks to Siegmar Gross for reporting the failure. 394 + The plugin ABI has changed, this release will not load plugins 395 built against previous hwloc releases. 396 397 398Version 1.9.1 399------------- 400* Fix a crash when the PCI locality is invalid. Attach to the root object 401 instead. Thanks to Nicolas Denoyelle for reporting the issue. 402* Fix -f in lstopo manpage. Thanks to Jirka Hladky for reporting the issue. 403* Fix hwloc_obj_type_sscanf() and others when strncasecmp() is not properly 404 available. Thanks to Nick Papior Andersen for reporting the problem. 405* Mark Linux file descriptors as close-on-exec to avoid leaks on exec. 406* Fix some minor memory leaks. 407 408 409Version 1.9.0 410------------- 411* API 412 + Add hwloc_obj_type_sscanf() to extend hwloc_obj_type_of_string() with 413 type-specific attributes such as Cache/Group depth and Cache type. 414 hwloc_obj_type_of_string() is moved to hwloc/deprecated.h. 415 + Add hwloc_linux_get_tid_last_cpu_location() for retrieving the 416 last CPU where a Linux thread given by TID ran. 417 + Add hwloc_distrib() to extend the old hwloc_distribute[v]() functions. 418 hwloc_distribute[v]() is moved to hwloc/deprecated.h. 419 + Don't mix total and local memory when displaying verbose object attributes 420 with hwloc_obj_attr_snprintf() or in lstopo. 421* Backends 422 + Add CPUVendor, CPUModelNumber and CPUFamilyNumber info attributes for 423 x86, ia64 and Xeon Phi sockets on Linux, to extend the x86-specific 424 support added in v1.8.1. Requested by Ralph Castain. 425 + Add many CPU- and Platform-related info attributes on ARM and POWER 426 platforms, in the Machine and Socket objects. 427 + Add CUDA info attributes describing the number of multiprocessors and 428 cores and the size of the global, shared and L2 cache memories in CUDA 429 OS devices. 430 + Add OpenCL info attributes describing the number of compute units and 431 the global memory size in OpenCL OS devices. 432 + The synthetic backend now accepts extended types such as L2Cache, L1i or 433 Group3. lstopo also exports synthetic strings using these extended types. 434* Tools 435 + lstopo 436 - Do not overwrite output files by default anymore. 437 Pass -f or --force to enforce it. 438 - Display OpenCL, CUDA and Xeon Phi numbers of cores and memory sizes 439 in the graphical output. 440 - Fix export to stdout when specifying a Cairo-based output type 441 with --of. 442 + hwloc-ps 443 - Add -e or --get-last-cpu-location to report where processes/threads 444 run instead of where they are bound. 445 - Report locations as likely-more-useful objects such as Cores or Sockets 446 instead of Caches when possible. 447 + hwloc-bind 448 - Fix failure on Windows when not using --pid. 449 - Add -e as a synonym to --get-last-cpu-location. 450 + hwloc-distrib 451 - Add --reverse to distribute using last objects first and singlify 452 into last bits first. Thanks to Jirka Hladky for the suggestion. 453 + hwloc-info 454 - Report unified caches when looking for data or instruction cache 455 ancestor objects. 456* Misc 457 + Add experimental Visual Studio support under contrib/windows. 458 Thanks to Eloi Gaudry for his help and for providing the first draft. 459 + Fix some overzealous assertions and warnings about the ordering of 460 objects on a level with respect to cpusets. The ordering is only 461 guaranteed for complete cpusets (based on the first bit in sets). 462 + Fix some memory leaks when importing xml diffs and when exporting a 463 "too complex" entry. 464 465 466Version 1.8.1 467------------- 468* Fix the cpuid code on Windows 64bits so that the x86 backend gets 469 enabled as expected and can populate CPU information. 470 Thanks to Robin Scher for reporting the problem. 471* Add CPUVendor/CPUModelNumber/CPUFamilyNumber attributes when running 472 on x86 architecture. Thanks to Ralph Castain for the suggestion. 473* Work around buggy BIOS reporting duplicate NUMA nodes on Linux. 474 Thanks to Jeff Becker for reporting the problem and testing the patch. 475* Add a name to the lstopo graphical window. Thanks to Michael Prokop 476 for reporting the issue. 477 478 479Version 1.8.0 480------------- 481* New components 482 + Add the "linuxpci" component that always works on Linux even when 483 libpciaccess and libpci aren't available (and even with a modified 484 file-system root). By default the old "pci" component runs first 485 because "linuxpci" lacks device names (obj->name is always NULL). 486* API 487 + Add the topology difference API in hwloc/diff.h for manipulating 488 many similar topologies. 489 + Add hwloc_topology_dup() for duplicating an entire topology. 490 + hwloc.h and hwloc/helper.h have been reorganized to clarify the 491 documentation sections. The actual inline code has moved out of hwloc.h 492 into the new hwloc/inlines.h. 493 + Deprecated functions are now in hwloc/deprecated.h, and not in the 494 official documentation anymore. 495* Tools 496 + Add hwloc-diff and hwloc-patch tools together with the new diff API. 497 + Add hwloc-compress-dir to (de)compress an entire directory of XML files 498 using hwloc-diff and hwloc-patch. 499 + Object colors in the graphical output of lstopo may be changed by adding 500 a "lstopoStyle" info attribute. See CUSTOM COLORS in the lstopo(1) manpage 501 for details. Thanks to Jirka Hladky for discussing the idea. 502 + hwloc-gather-topology may now gather I/O-related files on Linux when 503 --io is given. Only the linuxpci component supports discovering I/O 504 objects from these extended tarballs. 505 + hwloc-annotate now supports --ri to remove/replace info attributes with 506 a given name. 507 + hwloc-info supports "root" and "all" special locations for dumping 508 information about the root object. 509 + lstopo now supports --append-legend to append custom lines of text 510 to the legend in the graphical output. Thanks to Jirka Hladky for 511 discussing the idea. 512 + hwloc-calc and friends have a more robust parsing of locations given 513 on the command-line and they report useful error messages about it. 514 + Add --whole-system to hwloc-bind, hwloc-calc, hwloc-distances and 515 hwloc-distrib, and add --restrict to hwloc-bind for uniformity among 516 tools. 517* Misc 518 + Calling hwloc_topology_load() or hwloc_topology_set_*() on an already 519 loaded topology now returns an error (deprecated since release 1.6.1). 520 + Fix the initialisation of cpusets and nodesets in Group objects added 521 when inserting PCI hostbridges. 522 + Never merge Group objects that were added explicitly by the user with 523 hwloc_custom_insert_group_object_by_parent(). 524 + Add a sanity check during dynamic plugin loading to prevent some 525 crashes when hwloc is dynamically loaded by another plugin mechanisms. 526 + Add --with-hwloc-plugins-path to specify the install/load directories 527 of plugins. 528 + Add the MICSerialNumber info attribute to the root object when running 529 hwloc inside a Xeon Phi to match the same attribute in the MIC OS device 530 when running in the host. 531 532 533Version 1.7.2 534------------- 535* Do not create invalid block OS devices on very old Linux kernel such 536 as RHEL4 2.6.9. 537* Fix PCI subvendor/device IDs. 538* Fix the management of Misc objects inserted by parent. 539 Thanks to Jirka Hladky for reporting the problem. 540* Add a Port<n>State into attribute to OpenFabrics OS devices. 541* Add a MICSerialNumber info attribute to Xeon PHI/MIC OS devices. 542* Improve verbose error messages when failing to load from XML. 543 544 545Version 1.7.1 546------------- 547* Fix a failed assertion in the distance grouping code when loading a XML 548 file that already contains some groups. 549 Thanks to Laercio Lima Pilla for reporting the problem. 550* Remove unexpected Group objects when loading XML topologies with I/O 551 objects and NUMA distances. 552 Thanks to Elena Elkina for reporting the problem and testing patches. 553* Fix PCI link speed discovery when using libpciaccess. 554* Fix invalid libpciaccess virtual function device/vendor IDs when using 555 SR-IOV PCI devices on Linux. 556* Fix GL component build with old NVCtrl releases. 557 Thanks to Jirka Hladky for reporting the problem. 558* Fix embedding breakage caused by libltdl. 559 Thanks to Pavan Balaji for reporting the problem. 560* Always use the system-wide libltdl instead of shipping one inside hwloc. 561* Document issues when enabling plugins while embedding hwloc in another 562 project, in the documentation section Embedding hwloc in Other Software. 563* Add a FAQ entry "How to get useful topology information on NetBSD?" 564 in the documentation. 565* Somes fixes in the renaming code for embedding. 566* Miscellaneous minor build fixes. 567 568 569Version 1.7.0 570------------- 571* New operating system backends 572 + Add BlueGene/Q compute node kernel (CNK) support. See the FAQ in the 573 documentation for details. Thanks to Jeff Hammond, Christopher Samuel 574 and Erik Schnetter for their help. 575 + Add NetBSD support, thanks to Aleksej Saushev. 576* New I/O device discovery 577 + Add co-processor OS devices such as "mic0" for Intel Xeon Phi (MIC) 578 on Linux. Thanks to Jerome Vienne for helping. 579 + Add co-processor OS devices such as "cuda0" for NVIDIA CUDA-capable GPUs. 580 + Add co-processor OS devices such as "opencl0d0" for OpenCL GPU devices 581 on the AMD OpenCL implementation. 582 + Add GPU OS devices such as ":0.0" for NVIDIA X11 displays. 583 + Add GPU OS devices such as "nvml0" for NVIDIA GPUs. 584 Thanks to Marwan Abdellah and Stefan Eilemann for helping. 585 These new OS devices have some string info attributes such as CoProcType, 586 GPUModel, etc. to better identify them. 587 See the I/O Devices and Attributes documentation sections for details. 588* New components 589 + Add the "opencl", "cuda", "nvml" and "gl" components for I/O device 590 discovery. 591 + "nvml" also improves the discovery of NVIDIA GPU PCIe link speed. 592 All of these new components may be built as plugins. They may also be 593 disabled entirely by passing --disable-opencl/cuda/nvml/gl to configure. 594 See the I/O Devices, Components and Plugins, and FAQ documentation 595 sections for details. 596* API 597 + Add hwloc_topology_get_flags(). 598 + Add hwloc/plugins.h for building external plugins. 599 See the Adding new discovery components and plugins section. 600* Interoperability 601 + Add hwloc/opencl.h, hwloc/nvml.h, hwloc/gl.h and hwloc/intel-mic.h 602 to retrieve the locality of OS devices that correspond to AMD OpenCL 603 GPU devices or indexes, to NVML devices or indexes, to NVIDIA X11 604 displays, or to Intel Xeon Phi (MIC) device indexes. 605 + Add new helpers in hwloc/cuda.h and hwloc/cudart.h to convert 606 between CUDA devices or indexes and hwloc OS devices. 607 + Add hwloc_ibv_get_device_osdev() and clarify the requirements 608 of the OpenFabrics Verbs helpers in hwloc/openfabrics-verbs.h. 609* Tools 610 + hwloc-info is not only a synonym of lstopo -s anymore, it also 611 dumps information about objects given on the command-line. 612* Documentation 613 + Add a section "Existing components and plugins". 614 + Add a list of common OS devices in section "Software devices". 615 + Add a new FAQ entry "Why is lstopo slow?" about lstopo slowness 616 issues because of GPUs. 617 + Clarify the documentation of inline helpers in hwloc/myriexpress.h 618 and hwloc/openfabrics-verbs.h. 619* Misc 620 + Improve cache detection on AIX. 621 + The HWLOC_COMPONENTS variable now excludes the components whose 622 names are prefixed with '-'. 623 + lstopo --ignore PU now works when displaying the topology in 624 graphical and textual mode (not when exporting to XML). 625 + Make sure I/O options always appear in lstopo usage, not only when 626 using pciutils/libpci. 627 + Remove some unneeded Linux specific includes from some interoperability 628 headers. 629 + Fix some inconsistencies in hwloc-distrib and hwloc-assembler-remote 630 manpages. Thanks to Guy Streeter for the report. 631 + Fix a memory leak on AIX when getting memory binding. 632 + Fix many small memory leaks on Linux. 633 + The `libpci' component is now called `pci' but the old name is still 634 accepted in the HWLOC_COMPONENTS variable for backward compatibility. 635 636 637Version 1.6.2 638------------- 639* Use libpciaccess instead of pciutils/libpci by default for I/O discovery. 640 pciutils/libpci is only used if --enable-libpci is given to configure 641 because its GPL license may taint hwloc. See the Installation section 642 in the documentation for details. 643* Fix get_cpubind on Solaris when bound to a single PU with 644 processor_bind(). Thanks to Eugene Loh for reporting the problem 645 and providing a patch. 646 647 648Version 1.6.1 649------------- 650* Fix some crash or buggy detection in the x86 backend when Linux 651 cgroups/cpusets restrict the available CPUs. 652* Fix the pkg-config output with --libs --static. 653 Thanks to Erik Schnetter for reporting one of the problems. 654* Fix the output of hwloc-calc -H --hierarchical when using logical 655 indexes in the output. 656* Calling hwloc_topology_load() multiple times on the same topology 657 is officially deprecated. hwloc will warn in such cases. 658* Add some documentation about existing plugins/components, package 659 dependencies, and I/O devices specification on the command-line. 660 661 662Version 1.6.0 663------------- 664* Major changes 665 + Reorganize the backend infrastructure to support dynamic selection 666 of components and dynamic loading of plugins. For details, see the 667 new documentation section Components and plugins. 668 - The HWLOC_COMPONENTS variable lets one replace the default discovery 669 components. 670 - Dynamic loading of plugins may be enabled with --enable-plugins 671 (except on AIX and Windows). It will build libxml2 and libpci 672 support as separated modules. This helps reducing the dependencies 673 of the core hwloc library when distributed as a binary package. 674* Backends 675 + Add CPUModel detection on Darwin and x86/FreeBSD. 676 Thanks to Robin Scher for providing ways to implement this. 677 + The x86 backend now adds CPUModel info attributes to socket objects 678 created by other backends that do not natively support this attribute. 679 + Fix detection on FreeBSD in case of cpuset restriction. Thanks to 680 Sebastian Kuzminsky for reporting the problem. 681* XML 682 + Add hwloc_topology_set_userdata_import/export_callback(), 683 hwloc_export_obj_userdata() and _userdata_base64() to let 684 applications specify how to save/restore the custom data they placed 685 in the userdata private pointer field of hwloc objects. 686* Tools 687 + Add hwloc-annotate program to add string info attributes to XML 688 topologies. 689 + Add --pid-cmd to hwloc-ps to append the output of a command to each 690 PID line. May be used for showing Open MPI process ranks, see the 691 hwloc-ps(1) manpage for details. 692 + hwloc-bind now exits with an error if binding fails; the executable 693 is not launched unless binding suceeeded or --force was given. 694 + Add --quiet to hwloc-calc and hwloc-bind to hide non-fatal error 695 messages. 696 + Fix command-line pid support in windows tools. 697 + All programs accept --verbose as a synonym to -v. 698* Misc 699 + Fix some DIR descriptor leaks on Linux. 700 + Fix I/O device lists when some were filtered out after a XML import. 701 + Fix the removal of I/O objects when importing a I/O-enabled XML topology 702 without any I/O topology flag. 703 + When merging objects with HWLOC_IGNORE_TYPE_KEEP_STRUCTURE or 704 lstopo --merge, compare object types before deciding which one of two 705 identical object to remove (e.g. keep sockets in favor of caches). 706 + Add some GUID- and LID-related info attributes to OpenFabrics 707 OS devices. 708 + Only add CPUType socket attributes on Solaris/Sparc. Other cases 709 don't report reliable information (Solaris/x86), and a replacement 710 is available as the Architecture string info in the Machine object. 711 + Add missing Backend string info on Solaris in most cases. 712 + Document object attributes and string infos in a new Attributes 713 section in the documentation. 714 + Add a section about Synthetic topologies in the documentation. 715 716 717Version 1.5.2 (some of these changes are in v1.6.2 but not in v1.6) 718------------- 719* Use libpciaccess instead of pciutils/libpci by default for I/O discovery. 720 pciutils/libpci is only used if --enable-libpci is given to configure 721 because its GPL license may taint hwloc. See the Installation section 722 in the documentation for details. 723* Fix get_cpubind on Solaris when bound to a single PU with 724 processor_bind(). Thanks to Eugene Loh for reporting the problem 725 and providing a patch. 726* Fix some DIR descriptor leaks on Linux. 727* Fix I/O device lists when some were filtered out after a XML import. 728* Add missing Backend string info on Solaris in most cases. 729* Fix the removal of I/O objects when importing a I/O-enabled XML topology 730 without any I/O topology flag. 731* Fix the output of hwloc-calc -H --hierarchical when using logical 732 indexes in the output. 733* Fix the pkg-config output with --libs --static. 734 Thanks to Erik Schnetter for reporting one of the problems. 735 736 737Version 1.5.1 738------------- 739* Fix block OS device detection on Linux kernel 3.3 and later. 740 Thanks to Guy Streeter for reporting the problem and testing the fix. 741* Fix the cpuid code in the x86 backend (for FreeBSD). Thanks to 742 Sebastian Kuzminsky for reporting problems and testing patches. 743* Fix 64bit detection on FreeBSD. 744* Fix some corner cases in the management of the thissystem flag with 745 respect to topology flags and environment variables. 746* Fix some corner cases in command-line parsing checks in hwloc-distrib 747 and hwloc-distances. 748* Make sure we do not miss some block OS devices on old Linux kernels 749 when a single PCI device has multiple IDE hosts/devices behind it. 750* Do not disable I/O devices or instruction caches in hwloc-assembler output. 751 752 753Version 1.5.0 754------------- 755* Backends 756 + Do not limit the number of processors to 1024 on Solaris anymore. 757 + Gather total machine memory on FreeBSD. Thanks to Cyril Roelandt. 758 + XML topology files do not depend on the locale anymore. Float numbers 759 such as NUMA distances or PCI link speeds now always use a dot as a 760 decimal separator. 761 + Add instruction caches detection on Linux, AIX, Windows and Darwin. 762 + Add get_last_cpu_location() support for the current thread on AIX. 763 + Support binding on AIX when threads or processes were bound with 764 bindprocessor(). Thanks to Hendryk Bockelmann for reporting the issue 765 and testing patches, and to Farid Parpia for explaining the binding 766 interfaces. 767 + Improve AMD topology detection in the x86 backend (for FreeBSD) using 768 the topoext feature. 769* API 770 + Increase HWLOC_API_VERSION to 0x00010500 so that API changes may be 771 detected at build-time. 772 + Add a cache type attribute describind Data, Instruction and Unified 773 caches. Caches with different types but same depth (for instance L1d 774 and L1i) are placed on different levels. 775 + Add hwloc_get_cache_type_depth() to retrieve the hwloc level depth of 776 of the given cache depth and type, for instance L1i or L2. 777 It helps disambiguating the case where hwloc_get_type_depth() returns 778 HWLOC_TYPE_DEPTH_MULTIPLE. 779 + Instruction caches are ignored unless HWLOC_TOPOLOGY_FLAG_ICACHES is 780 passed to hwloc_topology_set_flags() before load. 781 + Add hwloc_ibv_get_device_osdev_by_name() OpenFabrics helper in 782 openfabrics-verbs.h to find the hwloc OS device object corresponding to 783 an OpenFabrics device. 784* Tools 785 + Add lstopo-no-graphics, a lstopo built without graphical support to 786 avoid dependencies on external libraries such as Cairo and X11. When 787 supported, graphical outputs are only available in the original lstopo 788 program. 789 - Packagers splitting lstopo and lstopo-no-graphics into different 790 packages are advised to use the alternatives system so that lstopo 791 points to the best available binary. 792 + Instruction caches are enabled in lstopo by default. Use --no-icaches 793 to disable them. 794 + Add -t/--threads to show threads in hwloc-ps. 795* Removal of obsolete components 796 + Remove the old cpuset interface (hwloc/cpuset.h) which is deprecated and 797 superseded by the bitmap API (hwloc/bitmap.h) since v1.1. 798 hwloc_cpuset and nodeset types are still defined, but all hwloc_cpuset_* 799 compatibility wrappers are now gone. 800 + Remove Linux libnuma conversion helpers for the deprecated and 801 broken nodemask_t interface. 802 + Remove support for "Proc" type name, it was superseded by "PU" in v1.0. 803 + Remove hwloc-mask symlinks, it was replaced by hwloc-calc in v1.0. 804* Misc 805 + Fix PCIe 3.0 link speed computation. 806 + Non-printable characters are dropped from strings during XML export. 807 + Fix importing of escaped characters with the minimalistic XML backend. 808 + Assert hwloc_is_thissystem() in several I/O related helpers. 809 + Fix some memory leaks in the x86 backend for FreeBSD. 810 + Minor fixes to ease native builds on Windows. 811 + Limit the number of retries when operating on all threads within a 812 process on Linux if the list of threads is heavily getting modified. 813 814 815Version 1.4.3 816------------- 817* This release is only meant to fix the pciutils license issue when upgrading 818 to hwloc v1.5 or later is not possible. It contains several other minor 819 fixes but ignores many of them that are only in v1.5 or later. 820* Use libpciaccess instead of pciutils/libpci by default for I/O discovery. 821 pciutils/libpci is only used if --enable-libpci is given to configure 822 because its GPL license may taint hwloc. See the Installation section 823 in the documentation for details. 824* Fix PCIe 3.0 link speed computation. 825* Fix importing of escaped characters with the minimalistic XML backend. 826* Fix a memory leak in the x86 backend. 827 828 829Version 1.4.2 830------------- 831* Fix build on Solaris 9 and earlier when fabsf() is not a compiler 832 built-in. Thanks to Igor Galić for reporting the problem. 833* Fix support for more than 32 processors on Windows. Thanks to Hartmut 834 Kaiser for reporting the problem. 835* Fix process-wide binding and cpulocation routines on Linux when some 836 threads disappear in the meantime. Thanks to Vlad Roubtsov for reporting 837 the issue. 838* Make installed scripts executable. Thanks to Jirka Hladky for reporting 839 the problem. 840* Fix libtool revision management when building for Windows. This fix was 841 also released as hwloc v1.4.1.1 Windows builds. Thanks to Hartmut Kaiser 842 for reporting the problem. 843* Fix the __hwloc_inline keyword in public headers when compiling with a 844 C++ compiler. 845* Add Port info attribute to network OS devices inside OpenFabrics PCI 846 devices so as to identify which interface corresponds to which port. 847* Document requirements for interoperability helpers: I/O devices discovery 848 is required for some of them; the topology must match the current host 849 for most of them. 850 851 852Version 1.4.1 853------------- 854* This release contains all changes from v1.3.2. 855* Fix hwloc_alloc_membind, thanks Karl Napf for reporting the issue. 856* Fix memory leaks in some get_membind() functions. 857* Fix helpers converting from Linux libnuma to hwloc (hwloc/linux-libnuma.h) 858 in case of out-of-order NUMA node ids. 859* Fix some overzealous assertions in the distance grouping code. 860* Workaround BIOS reporting empty I/O locality in CUDA and OpenFabrics 861 helpers on Linux. Thanks to Albert Solernou for reporting the problem. 862* Install a valgrind suppressions file hwloc-valgrind.supp (see the FAQ). 863* Fix memory binding documentation. Thanks to Karl Napf for reporting the 864 issues. 865 866 867Version 1.4.0 (does not contain all v1.3.2 changes) 868------------- 869* Major features 870 + Add "custom" interface and "assembler" tools to build multi-node 871 topology. See the Multi-node Topologies section in the documentation 872 for details. 873* Interface improvements 874 + Add symmetric_subtree object attribute to ease assumptions when consulting 875 regular symmetric topologies. 876 + Add a CPUModel and CPUType info attribute to Socket objects on Linux 877 and Solaris. 878 + Add hwloc_get_obj_index_inside_cpuset() to retrieve the "logical" index 879 of an object within a subtree of the topology. 880 + Add more NVIDIA CUDA helpers in cuda.h and cudart.h to find hwloc objects 881 corresponding to CUDA devices. 882* Discovery improvements 883 + Add a group object above partial distance matrices to make sure 884 the matrices are available in the final topology, except when this 885 new object would contradict the existing hierarchy. 886 + Grouping by distances now also works when loading from XML. 887 + Fix some corner cases in object insertion, for instance when dealing 888 with NUMA nodes without any CPU. 889* Backends 890 + Implement hwloc_get_area_membind() on Linux. 891 + Honor I/O topology flags when importing from XML. 892 + Further improve XML-related error checking and reporting. 893 + Hide synthetic topology error messages unless HWLOC_SYNTHETIC_VERBOSE=1. 894* Tools 895 + Add synthetic exporting of symmetric topologies to lstopo. 896 + lstopo --horiz and --vert can now be applied to some specific object types. 897 + lstopo -v -p now displays distance matrices with physical indexes. 898 + Add hwloc-distances utility to list distances. 899* Documentation 900 + Fix and/or document the behavior of most inline functions in hwloc/helper.h 901 when the topology contains some I/O or Misc objects. 902 + Backend documentation enhancements. 903* Bug fixes 904 + Fix missing last bit in hwloc_linux_get_thread_cpubind(). 905 Thanks to Carolina Gómez-Tostón Gutiérrez for reporting the issue. 906 + Fix FreeBSD build without cpuid support. 907 + Fix several Windows build issues. 908 + Fix inline keyword definition in public headers. 909 + Fix dependencies in the embedded library. 910 + Improve visibility support detection. Thanks to Dave Love for providing 911 the patch. 912 + Remove references to internal symbols in the tools. 913 914 915Version 1.3.3 916------------- 917* This release is only meant to fix the pciutils license issue when upgrading 918 to hwloc v1.4 or later is not possible. It contains several other minor 919 fixes but ignores many of them that are only in v1.4 or later. 920* Use libpciaccess instead of pciutils/libpci by default for I/O discovery. 921 pciutils/libpci is only used if --enable-libpci is given to configure 922 because its GPL license may taint hwloc. See the Installation section 923 in the documentation for details. 924 925 926Version 1.3.2 927------------- 928* Fix missing last bit in hwloc_linux_get_thread_cpubind(). 929 Thanks to Carolina Gómez-Tostón Gutiérrez for reporting the issue. 930* Fix build with -mcmodel=medium. Thanks to Devendar Bureddy for reporting 931 the issue. 932* Fix build with Solaris Studio 12 compiler when XML is disabled. 933 Thanks to Paul H. Hargrove for reporting the problem. 934* Fix installation with old GNU sed, for instance on Red Hat 8. 935 Thanks to Paul H. Hargrove for reporting the problem. 936* Fix PCI locality when Linux cgroups restrict the available CPUs. 937* Fix floating point issue when grouping by distance on mips64 architecture. 938 Thanks to Paul H. Hargrove for reporting the problem. 939* Fix conversion from/to Linux libnuma when some NUMA nodes have no memory. 940* Fix support for gccfss compilers with broken ffs() support. Thanks to 941 Paul H. Hargrove for reporting the problem and providing a patch. 942* Fix FreeBSD build without cpuid support. 943* Fix several Windows build issues. 944* Fix inline keyword definition in public headers. 945* Fix dependencies in the embedded library. 946* Detect when a compiler such as xlc may not report compile errors 947 properly, causing some configure checks to be wrong. Thanks to 948 Paul H. Hargrove for reporting the problem and providing a patch. 949* Improve visibility support detection. Thanks to Dave Love for providing 950 the patch. 951* Remove references to internal symbols in the tools. 952* Fix installation on systems with limited command-line size. 953 Thanks to Paul H. Hargrove for reporting the problem. 954* Further improve XML-related error checking and reporting. 955 956 957Version 1.3.1 958------------- 959* Fix pciutils detection with pkg-config when not installed in standard 960 directories. 961* Fix visibility options detection with the Solaris Studio compiler. 962 Thanks to Igor Galić and Terry Dontje for reporting the problems. 963* Fix support for old Linux sched.h headers such as those found 964 on Red Hat 8. Thanks to Paul H. Hargrove for reporting the problems. 965* Fix inline and attribute support for Solaris compilers. Thanks to 966 Dave Love for reporting the problems. 967* Print a short summary at the end of the configure output. Thanks to 968 Stefan Eilemann for the suggestion. 969* Add --disable-libnuma configure option to disable libnuma-based 970 memory binding support on Linux. Thanks to Rayson Ho for the 971 suggestion. 972* Make hwloc's configure script properly obey $PKG_CONFIG. Thanks to 973 Nathan Phillip Brink for raising the issue. 974* Silence some harmless pciutils warnings, thanks to Paul H. Hargrove 975 for reporting the problem. 976* Fix the documentation with respect to hwloc_pid_t and hwloc_thread_t 977 being either pid_t and pthread_t on Unix, or HANDLE on Windows. 978 979 980Version 1.3.0 981------------- 982* Major features 983 + Add I/O devices and bridges to the topology using the pciutils 984 library. Only enabled after setting the relevant flag with 985 hwloc_topology_set_flags() before hwloc_topology_load(). See the 986 I/O Devices section in the documentation for details. 987* Discovery improvements 988 + Add associativity to the cache attributes. 989 + Add support for s390/z11 "books" on Linux. 990 + Add the HWLOC_GROUPING_ACCURACY environment variable to relax 991 distance-based grouping constraints. See the Environment Variables 992 section in the documentation for details about grouping behavior 993 and configuration. 994 + Allow user-given distance matrices to remove or replace those 995 discovered by the OS backend. 996* XML improvements 997 + XML is now always supported: a minimalistic custom import/export 998 code is used when libxml2 is not available. It is only guaranteed 999 to read XML files generated by hwloc. 1000 + hwloc_topology_export_xml() and export_xmlbuffer() now return an 1001 integer. 1002 + Add hwloc_free_xmlbuffer() to free the buffer allocated by 1003 hwloc_topology_export_xmlbuffer(). 1004 + Hide XML topology error messages unless HWLOC_XML_VERBOSE=1. 1005* Minor API updates 1006 + Add hwloc_obj_add_info to customize object info attributes. 1007* Tools 1008 + lstopo now displays I/O devices by default. Several options are 1009 added to configure the I/O discovery. 1010 + hwloc-calc and hwloc-bind now accept I/O devices as input. 1011 + Add --restrict option to hwloc-calc and hwloc-distribute. 1012 + Add --sep option to change the output field separator in hwloc-calc. 1013 + Add --whole-system option to hwloc-ps. 1014 1015 1016Version 1.2.2 1017------------- 1018* Fix build on AIX 5.2, thanks Utpal Kumar Ray for the report. 1019* Fix XML import of very large page sizes or counts on 32bits platform, 1020 thanks to Karsten Hopp for the RedHat ticket. 1021* Fix crash when administrator limitations such as Linux cgroup require 1022 to restrict distance matrices. Thanks to Ake Sandgren for reporting the 1023 problem. 1024* Fix the removal of objects such as AMD Magny-Cours dual-node sockets 1025 in case of administrator restrictions. 1026* Improve error reporting and messages in case of wrong synthetic topology 1027 description. 1028* Several other minor internal fixes and documentation improvements. 1029 1030 1031Version 1.2.1 1032------------- 1033* Improve support of AMD Bulldozer "Compute-Unit" modules by detecting 1034 logical processors with different core IDs on Linux. 1035* Fix hwloc-ps crash when listing processes from another Linux cpuset. 1036 Thanks to Carl Smith for reporting the problem. 1037* Fix build on AIX and Solaris. Thanks to Carl Smith and Andreas Kupries 1038 for reporting the problems. 1039* Fix cache size detection on Darwin. Thanks to Erkcan Özcan for reporting 1040 the problem. 1041* Make configure fail if --enable-xml or --enable-cairo is given and 1042 proper support cannot be found. Thanks to Andreas Kupries for reporting 1043 the XML problem. 1044* Fix spurious L1 cache detection on AIX. Thanks to Hendryk Bockelmann 1045 for reporting the problem. 1046* Fix hwloc_get_last_cpu_location(THREAD) on Linux. Thanks to Gabriele 1047 Fatigati for reporting the problem. 1048* Fix object distance detection on Solaris. 1049* Add pthread_self weak symbol to ease static linking. 1050* Minor documentation fixes. 1051 1052 1053Version 1.2.0 1054------------- 1055* Major features 1056 + Expose latency matrices in the API as an array of distance structures 1057 within objects. Add several helpers to find distances. 1058 + Add hwloc_topology_set_distance_matrix() and environment variables 1059 to provide a matrix of distances between a given set of objects. 1060 + Add hwloc_get_last_cpu_location() and hwloc_get_proc_last_cpu_location() 1061 to retrieve the processors where a process or thread recently ran. 1062 - Add the corresponding --get-last-cpu-location option to hwloc-bind. 1063 + Add hwloc_topology_restrict() to restrict an existing topology to a 1064 given cpuset. 1065 - Add the corresponding --restrict option to lstopo. 1066* Minor API updates 1067 + Add hwloc_bitmap_list_sscanf/snprintf/asprintf to convert between bitmaps 1068 and strings such as 4-5,7-9,12,15- 1069 + hwloc_bitmap_set/clr_range() now support infinite ranges. 1070 + Clarify the difference between inserting Misc objects by cpuset or by 1071 parent. 1072 + hwloc_insert_misc_object_by_cpuset() now returns NULL in case of error. 1073* Discovery improvements 1074 + x86 backend (for freebsd): add x2APIC support 1075 + Support standard device-tree phandle, to get better support on e.g. ARM 1076 systems providing it. 1077 + Detect cache size on AIX. Thanks Christopher and IBM. 1078 + Improve grouping to support asymmetric topologies. 1079* Tools 1080 + Command-line tools now support "all" and "root" special locations 1081 consisting in the entire topology, as well as type names with depth 1082 attributes such as L2 or Group4. 1083 + hwloc-calc improvements: 1084 - Add --number-of/-N option to report the number of objects of a given 1085 type or depth. 1086 - -I is now equivalent to --intersect for listing the indexes of 1087 objects of a given type or depth that intersects the input. 1088 - Add -H to report the output as a hierarchical combination of types 1089 and depths. 1090 + Add --thissystem to lstopo. 1091 + Add lstopo-win, a console-less lstopo variant on Windows. 1092* Miscellaneous 1093 + Remove C99 usage from code base. 1094 + Rename hwloc-gather-topology.sh into hwloc-gather-topology 1095 + Fix AMD cache discovery on freebsd when there is no L3 cache, thanks 1096 Andriy Gapon for the fix. 1097 1098 1099Version 1.1.2 1100------------- 1101* Fix a segfault in the distance-based grouping code when some objects 1102 are not placed in any group. Thanks to Bernd Kallies for reporting 1103 the problem and providing a patch. 1104* Fix the command-line parsing of hwloc-bind --mempolicy interleave. 1105 Thanks to Guy Streeter for reporting the problem. 1106* Stop truncating the output in hwloc_obj_attr_snprintf() and in the 1107 corresponding lstopo output. Thanks to Guy Streeter for reporting the 1108 problem. 1109* Fix object levels ordering in synthetic topologies. 1110* Fix potential incoherency between device tree and kernel information, 1111 when SMT is disabled on Power machines. 1112* Fix and document the behavior of hwloc_topology_set_synthetic() in case 1113 of invalid argument. Thanks to Guy Streeter for reporting the problem. 1114* Add some verbose error message reporting when it looks like the OS 1115 gives erroneous information. 1116* Do not include unistd.h and stdint.h in public headers on Windows. 1117* Move config.h files into their own subdirectories to avoid name 1118 conflicts when AC_CONFIG_HEADERS adds -I's for them. 1119* Remove the use of declaring variables inside "for" loops. 1120* Some other minor fixes. 1121* Many minor documentation fixes. 1122 1123 1124Version 1.1.1 1125------------- 1126* Add hwloc_get_api_version() which returns the version of hwloc used 1127 at runtime. Thanks to Guy Streeter for the suggestion. 1128* Fix the number of hugepages reported for NUMA nodes on Linux. 1129* Fix hwloc_bitmap_to_ulong() right after allocating the bitmap. 1130 Thanks to Bernd Kallies for reporting the problem. 1131* Fix hwloc_bitmap_from_ith_ulong() to properly zero the first ulong. 1132 Thanks to Guy Streeter for reporting the problem. 1133* Fix hwloc_get_membind_nodeset() on Linux. 1134 Thanks to Bernd Kallies for reporting the problem and providing a patch. 1135* Fix some file descriptor leaks in the Linux discovery. 1136* Fix the minimum width of NUMA nodes, caches and the legend in the graphical 1137 lstopo output. Thanks to Jirka Hladky for reporting the problem. 1138* Various fixes to bitmap conversion from/to taskset-strings. 1139* Fix and document snprintf functions behavior when the buffer size is too 1140 small or zero. Thanks to Guy Streeter for reporting the problem. 1141* Fix configure to avoid spurious enabling of the cpuid backend. 1142 Thanks to Tim Anderson for reporting the problem. 1143* Cleanup error management in hwloc-gather-topology.sh. 1144 Thanks to Jirka Hladky for reporting the problem and providing a patch. 1145* Add a manpage and usage for hwloc-gather-topology.sh on Linux. 1146 Thanks to Jirka Hladky for providing a patch. 1147* Memory binding documentation enhancements. 1148 1149 1150Version 1.1.0 1151------------- 1152 1153* API 1154 + Increase HWLOC_API_VERSION to 0x00010100 so that API changes may be 1155 detected at build-time. 1156 + Add a memory binding interface. 1157 + The cpuset API (hwloc/cpuset.h) is now deprecated. It is replaced by 1158 the bitmap API (hwloc/bitmap.h) which offers the same features with more 1159 generic names since it applies to CPU sets, node sets and more. 1160 Backward compatibility with the cpuset API and ABI is still provided but 1161 it will be removed in a future release. 1162 Old types (hwloc_cpuset_t, ...) are still available as a way to clarify 1163 what kind of hwloc_bitmap_t each API function manipulates. 1164 Upgrading to the new API only requires to replace hwloc_cpuset_ function 1165 calls with the corresponding hwloc_bitmap_ calls, with the following 1166 renaming exceptions: 1167 - hwloc_cpuset_cpu -> hwloc_bitmap_only 1168 - hwloc_cpuset_all_but_cpu -> hwloc_bitmap_allbut 1169 - hwloc_cpuset_from_string -> hwloc_bitmap_sscanf 1170 + Add an `infos' array in each object to store couples of info names and 1171 values. It enables generic storage of things like the old dmi board infos 1172 that were previously stored in machine specific attributes. 1173 + Add linesize cache attribute. 1174* Features 1175 + Bitmaps (and thus CPU sets and node sets) are dynamically (re-)allocated, 1176 the maximal number of CPUs (HWLOC_NBMAXCPUS) has been removed. 1177 + Improve the distance-based grouping code to better support irregular 1178 distance matrices. 1179 + Add support for device-tree to get cache information (useful on Power 1180 architectures). 1181* Helpers 1182 + Add NVIDIA CUDA helpers in cuda.h and cudart.h to ease interoperability 1183 with CUDA Runtime and Driver APIs. 1184 + Add Myrinet Express helper in myriexpress.h to ease interoperability. 1185* Tools 1186 + lstopo now displays physical/OS indexes by default in graphical mode 1187 (use -l to switch back to logical indexes). The textual output still uses 1188 logical by default (use -p to switch to physical indexes). 1189 + lstopo prefixes logical indexes with `L#' and physical indexes with `P#'. 1190 Physical indexes are also printed as `P#N' instead of `phys=N' within 1191 object attributes (in parentheses). 1192 + Add a legend at the bottom of the lstopo graphical output, use --no-legend 1193 to remove it. 1194 + Add hwloc-ps to list process' bindings. 1195 + Add --membind and --mempolicy options to hwloc-bind. 1196 + Improve tools command-line options by adding a generic --input option 1197 (and more) which replaces the old --xml, --synthetic and --fsys-root. 1198 + Cleanup lstopo output configuration by adding --output-format. 1199 + Add --intersect in hwloc-calc, and replace --objects with --largest. 1200 + Add the ability to work on standard input in hwloc-calc. 1201 + Add --from, --to and --at in hwloc-distrib. 1202 + Add taskset-specific functions and command-line tools options to 1203 manipulate CPU set strings in the format of the taskset program. 1204 + Install hwloc-gather-topology.sh on Linux. 1205 1206 1207Version 1.0.3 1208------------- 1209 1210* Fix support for Linux cpuset when emulated by a cgroup mount point. 1211* Remove unneeded runtime dependency on libibverbs.so in the library and 1212 all utils programs. 1213* Fix hwloc_cpuset_to_linux_libnuma_ulongs in case of non-linear OS-indexes 1214 for NUMA nodes. 1215* lstopo now displays physical/OS indexes by default in graphical mode 1216 (use -l to switch back to logical indexes). The textual output still uses 1217 logical by default (use -p to switch to physical indexes). 1218 1219 1220Version 1.0.2 1221------------- 1222 1223* Public headers can now be included directly from C++ programs. 1224* Solaris fix for non-contiguous cpu numbers. Thanks to Rolf vandeVaart for 1225 reporting the issue. 1226* Darwin 10.4 fix. Thanks to Olivier Cessenat for reporting the issue. 1227* Revert 1.0.1 patch that ignored sockets with unknown ID values since it 1228 only slightly helped POWER7 machines with old Linux kernels while it 1229 prevents recent kernels from getting the complete POWER7 topology. 1230* Fix hwloc_get_common_ancestor_obj(). 1231* Remove arch-specific bits in public headers. 1232* Some fixes in the lstopo graphical output. 1233* Various man page clarifications and minor updates. 1234 1235 1236Version 1.0.1 1237------------- 1238 1239* Various Solaris fixes. Thanks to Yannick Martin for reporting the issue. 1240* Fix "non-native" builds on x86 platforms (e.g., when building 32 1241 bit executables with compilers that natively build 64 bit). 1242* Ignore sockets with unknown ID values (which fixes issues on POWER7 1243 machines). Thanks to Greg Bauer for reporting the issue. 1244* Various man page clarifications and minor updates. 1245* Fixed memory leaks in hwloc_setup_group_from_min_distance_clique(). 1246* Fix cache type filtering on MS Windows 7. Thanks to Αλέξανδρος 1247 Παπαδογιαννάκ for reporting the issue. 1248* Fixed warnings when compiling with -DNDEBUG. 1249 1250 1251Version 1.0.0 1252------------- 1253 1254* The ABI of the library has changed. 1255* Backend updates 1256 + Add FreeBSD support. 1257 + Add x86 cpuid based backend. 1258 + Add Linux cgroup support to the Linux cpuset code. 1259 + Support binding of entire multithreaded process on Linux. 1260 + Fix and enable Group support in Windows. 1261 + Cleanup XML export/import. 1262* Objects 1263 + HWLOC_OBJ_PROC is renamed into HWLOC_OBJ_PU for "Processing Unit", 1264 its stringified type name is now "PU". 1265 + Use new HWLOC_OBJ_GROUP objects instead of MISC when grouping 1266 objects according to NUMA distances or arbitrary OS aggregation. 1267 + Rework memory attributes. 1268 + Add different cpusets in each object to specify processors that 1269 are offline, unavailable, ... 1270 + Cleanup the storage of object names and DMI infos. 1271* Features 1272 + Add support for looking up specific PID topology information. 1273 + Add hwloc_topology_export_xml() to export the topology in a XML file. 1274 + Add hwloc_topology_get_support() to retrieve the supported features 1275 for the current topology context. 1276 + Support non-SYSTEM object as the root of the tree, use MACHINE in 1277 most common cases. 1278 + Add hwloc_get_*cpubind() routines to retrieve the current binding 1279 of processes and threads. 1280* API 1281 + Add HWLOC_API_VERSION to help detect the currently used API version. 1282 + Add missing ending "e" to *compare* functions. 1283 + Add several routines to emulate PLPA functions. 1284 + Rename and rework the cpuset and/or/xor/not/clear operators to output 1285 their result in a dedicated argument instead of modifying one input. 1286 + Deprecate hwloc_obj_snprintf() in favor of hwloc_obj_type/attr_snprintf(). 1287 + Clarify the use of parent and ancestor in the API, do not use father. 1288 + Replace hwloc_get_system_obj() with hwloc_get_root_obj(). 1289 + Return -1 instead of HWLOC_OBJ_TYPE_MAX in the API since the latter 1290 isn't public. 1291 + Relax constraints in hwloc_obj_type_of_string(). 1292 + Improve displaying of memory sizes. 1293 + Add 0x prefix to cpuset strings. 1294* Tools 1295 + lstopo now displays logical indexes by default, use --physical to 1296 revert back to OS/physical indexes. 1297 + Add colors in the lstopo graphical outputs to distinguish between online, 1298 offline, reserved, ... objects. 1299 + Extend lstopo to show cpusets, filter objects by type, ... 1300 + Renamed hwloc-mask into hwloc-calc which supports many new options. 1301* Documentation 1302 + Add a hwloc(7) manpage containing general information. 1303 + Add documentation about how to switch from PLPA to hwloc. 1304 + Cleanup the distributed documentation files. 1305* Miscellaneous 1306 + Many compilers warning fixes. 1307 + Cleanup the ABI by using the visibility attribute. 1308 + Add project embedding support. 1309 1310 1311Version 0.9.4 (unreleased) 1312-------------------------- 1313 1314* Fix reseting colors to normal in lstopo -.txt output. 1315* Fix Linux pthread_t binding error report. 1316 1317 1318Version 0.9.3 1319------------- 1320 1321* Fix autogen.sh to work with Autoconf 2.63. 1322* Fix various crashes in particular conditions: 1323 - xml files with root attributes 1324 - offline CPUs 1325 - partial sysfs support 1326 - unparseable /proc/cpuinfo 1327 - ignoring NUMA level while Misc level have been generated 1328* Tweak documentation a bit 1329* Do not require the pthread library for binding the current thread on Linux 1330* Do not erroneously consider the sched_setaffinity prototype is the old version 1331 when there is actually none. 1332* Fix _syscall3 compilation on archs for which we do not have the 1333 sched_setaffinity system call number. 1334* Fix AIX binding. 1335* Fix libraries dependencies: now only lstopo depends on libtermcap, fix 1336 binutils-gold link 1337* Have make check always build and run hwloc-hello.c 1338* Do not limit size of a cpuset. 1339 1340 1341Version 0.9.2 1342------------- 1343 1344* Trivial documentation changes. 1345 1346 1347Version 0.9.1 1348------------- 1349 1350* Re-branded to "hwloc" and moved to the Open MPI project, relicensed under the 1351 BSD license. 1352* The prefix of all functions and tools is now hwloc, and some public 1353 functions were also renamed for real. 1354* Group NUMA nodes into Misc objects according to their physical distance 1355 that may be reported by the OS/BIOS. 1356 May be ignored by setting HWLOC_IGNORE_DISTANCES=1 in the environment. 1357* Ignore offline CPUs on Solaris. 1358* Improved binding support on AIX. 1359* Add HP-UX support. 1360* CPU sets are now allocated/freed dynamically. 1361* Add command line options to tune the lstopo graphical output, add 1362 semi-graphical textual output 1363* Extend topobind to support multiple cpusets or objects on the command 1364 line as topomask does. 1365* Add an Infiniband-specific helper hwloc/openfabrics-verbs.h to retrieve 1366 the physical location of IB devices. 1367 1368 1369Version 0.9 (libtopology) 1370------------------------- 1371 1372* First release. 1373