Changes

Summary

1. doxy: more links from/to the topo attributes section (details)
2. completions: add info/calc/bind/distrib --restrict-flags (details)
3. contrib/dist/publish_doc: display home page links (details)
4. linux: add support for generic-initiator ACPI HMAT information (details)
5. Move PGI compiler check before GNU (details)
6. linux: fix comments about generic initiators (details)
7. linux: add support for generic-initiator ACPI HMAT performance information (details)
8. tests/linux: add a sysfs test with complex initiators/targets (details)
9. tests/linux: remove fakememinitiators-1npc+1npc (details)
10. contrib/android: fix the VERSION (details)
11. contrib/android: also fix version individual major/minor/release numbers (details)
12. contrib/android: also update the apk version (details)
13. contrib/android: remove HWLOC_VERSION from private config.h (details)
14. contrib/git-hooks: add a script for checking version numbers in pre-commit hooks (details)
15. contrib/git-hooks: add a pre-commit hook template (details)
16. bitmap: fix hwloc_bitmap_nr_ulongs() (details)
17. tests/bitmap: check bitmaps with single 0 bit and single full ulong (details)
18. contrib/ci.inria.fr: display the windows MSVC build environment (details)
19. contrib/ci.inria.fr: run hwloc-info --support on cygwin too (details)
20. lstopo/cairo: only set the line thickness once (details)
21. contrib/android: update to apk 1.5.1 (details)
22. contrib/android: cleanup RGB color parameter names (details)
23. git-hooks/check-version: update to new android gradle version format (details)
24. lstopo-ascii: Use last colors instead of first colors (details)
25. memattrs: try to clarify that most initiators should be cpusets for now (details)
26. memattrs: some doxygen fixes (details)
27. linux: factorize some fd/path reading functions back (details)
28. linux: rework read_path_by_length() return value (details)
29. linux: use helpers for reading tid location (details)
30. linux: use helpers for reading nvidia gpu numa nodes (details)
31. linux: use helpers for reading os device numa node (details)
32. linux: use helpers for reading the pci config space (details)
33. tests/diff: fix the test when the first NUMA node is CPU less (details)
34. core: abstract-out hwloc__tma_dup_infos() (details)
35. API: add hwloc/cpukinds.h for exposing difference between cores (details)
36. xml/import: factorize out the actual parsing on non-object-specific info attr (details)
37. xml: add cpukinds import/export (details)
38. tests/xml: add a case with cpukinds (details)
39. lstopo: add --cpukinds option (details)
40. x86: use the hybrid leaf to build a cpukinds based on the core type (details)
41. linux: read cpufreq/cpuinfo_max_freq and base_frequency to create cpukinds (details)
42. windows: use RtlGetVersion() for reliable version checks without manifest (details)
43. windows: add cpukinds using the native efficiency class in Windows 10 (details)
44. cpukinds: add some ranking/efficiency heuristics (details)
45. cpukinds: deduplicate identical info names/values (details)
46. doxy: add cpukinds info in topology attributes (details)
47. tests/linux: add heterogenous frequencies to fakeheteronuma for cpukinds (details)
48. tests/x86: add a test with fake cpukind information (details)
49. utils/hwloc-calc: add --cpukind (details)
50. utils/tests: check cpukinds in test-hwloc-calc (details)
51. utils/hwloc-info: display which CPU kind(s) CPU-side objects are included in (details)
52. utils/tests: check hwloc-info cpukinds output (details)
53. API: add hwloc_cpukinds_register() (details)
54. tests: add a cpukinds API test (details)
55. utils/hwloc-annotate: add cpukind command (details)
56. utils/tests: check cpukinds in hwloc-annotate (details)
57. API: add a cpukind_efficiency bit in the discovery support array (details)
58. lstopo/core: add cpukind attributes to object userdata (details)
59. lstopo/draw: change the format of PU boxes for CPU kinds (details)
60. linux/cpukinds: maxfreq isn't always maxfreq anymore (details)
61. linux/cpukinds: sort by frequency before registering (details)
62. API: bump HWLOC_API_VERSION to 0x00020400 (details)
63. ci.inria.fr/browse_jenkins_logs.sh: update examples (details)
64. contrib/android: fix the apk manifest name (details)
65. contrib/android: save the app icon original files (details)
66. contrib/android: fix sharing of graphic output (details)
67. linux: use Linux 5.11 sysfs cpu types to detect Intel Atom/Core types (details)
68. tests/linux: add Linux sysfs cpu types to fakeheterocpunuma (details)
69. doxy: fix LinuxCPUTyle cpukind attribute description (details)
70. VERSION: master is now 2.5 (details)
71. cpukinds: fix the checking for duplicate values when ranking by info (details)
72. distances: uniformize and document NUMALatency matrices (details)
73. autogen: fix MacOS X Big Sur libtool.m4 bug by patching configure (details)
74. utils/misc: remove a useless and harmless loop duplicate (details)
75. utils/misc: improve the printing of distance matrices (details)
76. core: fortify bridge downstream attribute use (details)
77. git-hooks/check-versions: better error messages when Windows VERSION is outdated (details)
78. NEWS: bring official 2.4.0 bullets in master (details)
79. Fixed a memory leak when destroying hwloc_linux_cpufreqs objects. (details)
80. distances: fix the import of large heterogeneous distance matrices (details)
81. tests/xml/fakeheterodistances: enlarge matrices to test hetero-indexes on multiple lines (details)
82. plugins.h: clarify that things may change when the component ABI is bumped (details)
83. ci.inria.fr/sonar: update languages for recent sonarscanner releases (details)
84. ci.inria.fr/sonar: output the sonar-scanner version (details)
85. configure: add --with-cuda=<dir> (details)
86. hwloc/opencl.h: fix AMD OpenCL device locality when PCI bus/device >= 128 (details)
87. ci.inria.fr/sonarscanner: update exclusions (details)
88. Fix GCC checks to use the vendor name (details)
89. contrib/android: remove unused duplicate android manifest (details)
90. x86: add a summary of PU/APIC/Core/Pkg/Node info in verbose debug (details)
91. x86: fix full discovery enabling when FreeBSD/NetBSD backends only created PUs (details)
92. x86: update general comments about this backend (details)
93. doc: Add missing dependency (details)
94. configure: pass -Werror when checking compiler options (details)
95. autogen: use bash explicitly (details)
96. configure: detect when a patched libtool was used (details)
97. core: mark the topology as modified when we replace an object (details)
98. core: fix inserting of custom groups replacing an existing ones (details)
99. tests/groups: check the merging of identical groups with higher/smaller kinds (details)
100. API: clarify the documentation of alloc/insert_group() (details)
101. lstopo: hide --top groups whenever possible (details)
102. core: add assertions about local/total memory in debug (details)
103. utils/hwloc-ps.1: fix a ref to --disallowed (details)
104. tests/shmem: cleanup the enabling of specific distance checks (details)
105. tests/shmem: also test topologies with memattrs/cpukinds/io from XML (details)
106. Fix check for IBM xl compilers for v13.1 and later. (details)
107. Fix configury where most compilers will get mislabeled as 'gnu'. (details)
108. utils/gather-cpuid: dump the 0x19 leaf (details)
109. configure: don't define cuda common vars unless needed (opencl/nvml/cuda) (details)
110. cpukinds: fix the checking for duplicate ranking values (details)
111. tests/x86+linux: allow .env file to change the x86,linux order (details)
112. tests/x86+linux: add a test combining x86 CoreType and linux cpufreq (details)
113. tests/x86+linux: don't forget the new options file in make dist (details)
114. linux: allow annotating cpukinds when PUs were discovered by another backend (x86) (details)
115. linux: fix support for CPU0 being hotplugged (details)
116. hwloc-ps: add --short-name (details)
117. hwloc-ps: don't apply --pid-cmd to processes that will be filtered out (details)
118. cuda/nvml/opencl/pci/rsmi: prefix error messages with "hwloc" (details)
119. utils/common-ps: fix popen/pclose for Windows (details)
120. configure: fix cuda common variable definition (details)
121. NEWS: bring 2.4.1 bullets in master (details)
122. utils/common-ps: refix popen/pclose for Windows/Cygwin (details)
123. configure: move AC_ARG_VAR(MS_LIB) to hwloc-internal config (details)
124. configure: typo in comments (details)
125. doxy: add Cygwin to the FAQ about Windows builds (details)
126. configure: cleanup cuda/cudart variables (details)
127. configure: use LIBS variables for ibverbs/rsmi/nvml tests build (details)
128. configure: use CPPFLAGS/CFLAGS/LIBS/LDFLAGS consistently (details)
129. configure: add --with-cuda-version or use CUDA_VERSION to select CUDA pkg-config files (details)
130. doxy: add a FAQ entry about CUDA configuration (details)
131. cuda: don't warn if no device were found (details)
132. API: add hwloc/windows.h for querying windows processor groups (details)
133. tests: add a test for the windows processor group API (details)
134. lstopo: add --windows-processor-groups (details)
135. doxy: add links from interoperability items to dedicated API sections (details)
136. windows: fix a leak in the initialization of processor group cpusets (details)
137. windows: properly destroy the processor group cpuset array (details)
138. lstopo: fix --windows-processor-groups name confusion (details)
139. contrib/ci.inria.fr: add --windows-processor-groups to windows lstopo outputs (details)
140. core: don't display "less critical" error messages by default (details)
141. components: cleanup error messages (details)
142. AIX: cleanup error messages (details)
143. solaris: cleanup error messages (details)
144. cpukinds: prefix the error message with "hwloc: " (details)
145. pci: cleanup some error messages (details)
146. linux/knl: cleanup error messages (details)
147. linux: cleanup more error messages (details)
148. linux+xml: cleanup some error messages about 32bits PCI domain (details)
149. xml: add a missing verbose condition before a printf (details)
150. xml: cleanup some error messages (details)
151. core: cleanup error messages (details)
152. NEWS: reorder items (details)
153. rsmi: also limit rsmi shutdown calls on error (details)
154. hwloc.ps: add --single-ancestor option (details)
155. contrib/completion: add lstopo --windows-processor-groups (details)
156. contrib/completion: add hwloc-ps --short-name (details)
157. doxy: add a missing /ul tag (details)
158. nvml/rsmi/gl: add NVML/RSMI/Display subtype of OS devices (details)
159. API: add hwloc_get_obj_with_same_locality() (details)
160. distances.h: typo in comments (details)
161. utils: display PCI bus IDs in distance matrices (details)
162. doxy: cite DRM card0/renderD128 devices in OS device examples (details)
163. OpenCL: the OpenCL device type is a bitset (details)
164. tests/ports/opencl: add missing CL_DEVICE_TYPE_DEFAULT (details)
165. linux: add veX osdevices for NEC Vector Engine cards (details)
166. lstopo/draw: render VE osdev memory and cores attributes (details)
167. gather-topology: gather the "ve" class for NEC vector engine sysfs attributes (details)
168. tests/linux: add a testcase with NEC Vector Engine (details)
169. cuda.h: cleanup return values in the doc (details)
170. cudart.h: cleanup return values in the doc (details)
171. gl.h: cleanup return values in the doc (details)
172. nvml.h: cleanup return values in the doc (details)
173. opencl.h: cleanup return values in the doc (details)
174. openfabrics-verbs.h: cleanup return values in the doc (details)
175. rsmi.h: cleanup return values in the doc (details)
176. doxy: specify that VectorEngine info attrs are for VectorEngine OS devices (details)
177. Add a "levelzero" backend (details)
178. config: check for __attribute__((constructor)) (details)
179. levelzero: define ZES_ENABLE_SYSMAN=1 in the library constructor and Windows DllMain (details)
180. levelzero: add hwloc/levelzero.h with helpers (details)
181. tests: add a levelzero test (details)
182. configure: add support for level-zero.pc (details)
183. tests/rename: pass rsmi/nvml/cuda/opencl/levelzero CPPFLAGS (details)
184. Make -ltermcap the last option for ncurses (details)
185. distances: fix the restricting of heterogeneous distances (details)
186. distances: add hwloc_distances_transform() (details)
187. tests/distances: test hwloc_distances_transform() (details)
188. distances.h: specify that we don't group by heterogeneous distances (details)
189. distances: move forward declarations together (details)
190. distances: rework internals of adding distances (details)
191. plugins.h: export low-level distances add_create/add_values/add_commit functions to plugins (details)
192. distances.h: replace add() with add_create(), add_values(), add_commit() (details)
193. doxy: split the distances_add section into add and remove sections (details)
194. tests: update to new distances add API (details)
195. utils/annotate: update to new distances add API (details)
196. utils/annotate: new distances may get a name now (details)
197. utils/annotate: cleanup/improve get_unique_obj() (details)
198. utils/annotate: add a new distances-transform command (details)
199. utils/annotate: tests distances-transform command (details)
200. pci: keep PCI-to-non-PCI bridges until the core filters them out (details)
201. plugins.h: clarify the documentation of hwloc_pci_find_parent_by_busid() (details)
202. plugins.h: export hwloc_pci_find_by_busid() to plugins (details)
203. nvml: add a NVLinkBandwidth matrix (details)
204. nvml: recognize and keep NVSwitches (details)
205. rsmi: add the XGMIBandwidth matrix (details)
206. doxy: talk about distances-transform in the GPU distances section (details)
207. distances: allow to add distances with the heterogeneous kind (details)
208. distances: one more ref to transform() in the comments (details)
209. utils/manpages: fix some escaping (details)
210. completions: add lstopo --thickness (details)
211. utils/gather-topology.1: add --keep (details)
212. hwloc-ps: --pid-cmd now supports "mpirank" and "env=<name>" (details)
213. utils: uniformize the basenam'ing of the argv[0] in usage (details)
214. utils/hwloc-annotate: add --version (details)
215. utils/hwloc-ps: add --version (details)
216. utils/hwloc-dump-hwdata: add --version (details)
217. utils/hwloc-gather-cpuid: add --version (details)
218. utils/hwloc-gather-topology: add --version (details)
219. utils/hwloc-compress-dir: add --version (details)
220. utils/hwloc-calc: properly support -h and --help (details)
221. utils: show -h --help in usage (details)
222. linux: fix support for NUMA node0 being offline (details)
223. tests/linux: add a testcase with CPU#0 and NUMA node#0 offline (details)
224. contrib/git-hooks: add a pre-commit hook checking options in manpage/completion (details)
225. lstopo.1: add a (harmless) missing \fR (details)
226. doc: add support for building the HTML doxygen on readthedocs.io (details)
227. levelzero: add completion queue groups information (details)
228. git-hooks/check-manpages-completions.pl: support git worktree (details)
229. hwloc2.dtd: add name attribute for distances2hetero (details)
230. utils/annotate: allow to replace distances objects with NULL to remove them (details)
231. distances: add MERGE_SWITCH_PORTS and TRANSITIVE_CLOSURE transformations (details)
232. utils/annotate: add merge-switch-ports and transitive-closure distances transformations (details)
233. lstopo: add --distances-transform for basic transformation (details)
234. tests/xml: add power8 distances between NVIDIA GPUs and Packages (details)
235. utils/annotate: add a verbose error message (details)
236. tests/annotate: check the replacing of cuda/opencl package/numa in GPU distances (details)
237. NEWS cleanups (details)
238. levelzero: fix device name numbering (details)
239. cpukinds: don't require both coretype+frequency by default (details)
240. cpukinds: remove arch-specific ranking strategy selection (details)
241. lstopo.desktop: add Icon=hwloc (details)
242. utils: add missing 0x in error messages about invalid flags (details)
243. levelzero: prefix headers include path with level_zero/ (details)
244. tests/ports: actually move levelzero headers to their new location (details)
245. level_zero: fix #include in test and helper (details)
246. lstopo/memattrs: remove a duplicate line-break on unrecognized initiator type (details)
247. lstopo/memattrs: fix the displaying of object initiators (details)
248. utils/calc: display unknown OS index as -1 (details)
249. utils/info: don't display unknown memattr initiator OS index (details)
250. lstopo/memattrs: cleanup object printing (details)
251. utils/hwloc-calc: Fix erroneous index in help message (details)
252. API: add HWLOC_TOPOLOGY_FLAG_RESTRICT_TO_{CPU,MEM}BINDING (details)
253. API: add HWLOC_TOPOLOGY_FLAG_DONT_CHANGE_BINDING (details)
254. contrib/windows: properly filter-in shmem.c (details)
255. doxy: improve XGMI/NVLink distances replacing/transforming examples (details)
256. doc: add a note about replacing objects in hwloc_distances_transform() (details)
257. configure: use AC_CHECK_HEADERS instead of AC_HAVE_HEADERS (details)
258. configure: use AS_HELP_STRING instead of AC_HELP_STRING (details)
259. configure: use AC_ENABLE/DISABLE_SHARED/STATIC instead of AM_ (details)
260. configure: use AC_LINK_IFELSE instead of AC_TRY_LINK (details)
261. configure: use AC_COMPILE_IFELSE instead of AC_TRY_COMPILE (details)
262. configure: don't call AC_PROG_CC_C99 on recent autoconf (details)
263. VERSION: master is now 2.6 (details)
264. shmem test: avoid using PATH_MAX inconditionally (details)
265. hwloc_get_obj_with_same_locality: avoid using PATH_MAX inconditionally (details)
266. windows: warn if we find processor groups on a 32bits windows (details)
267. API: real-time tasks may also require the RESTRICT_TO_BINDING flag (details)
268. rsmi: fix the XGMIPeers info attribute (details)
269. NEWS: update 2.5.0 news bullets (details)
270. lstopo/graphics: factorize the check for object cpukind styles (details)
271. doxy: update cpukinds ranking env var possible values (details)
272. cpukinds: try to clarify the doc about efficiency ranking (details)
273. utils/calc: add an example with --cpukind (details)
274. cpukinds: improve the documentation about native efficiencies (details)
275. darwin: add cpukinds support using IOKit cpus information (details)
276. linux/cpukinds: generalize the code to parse something else than cpufreq values (details)
277. linux: use the optional "cpu_capacity" sysfs cpu file to rank cpukinds (details)
278. linux/cpukinds: remove sysfs cpu type support (details)
279. tests/linux: update fakeheterocpunuma to use cpu_capacity instead of cpu_types (details)
280. pci: the HWLOC_PCI_LOCALITY file can contain comments (details)
281. tests/gather: ignore missing DAXDevice NUMA info attr (details)
282. lstopo/x11: fix autoresizing when window is maximized (details)
283. doxy/calc.1: more words about nodeset outputs in the hwloc-calc descriptions (details)
284. doxy+tools: more docs about cpukinds and ranking (details)
285. levelzero: fix the management of unknown PCIe bandwidth (details)
286. windows/dolib: print the entire lib command-line (details)
287. pci: PCIe Gen6 will likely be 64GT/s as expected (details)
288. lstopo.1: typo in --windows-processor-groups formatting (details)
289. configure: look for cuda in /usr/local/cuda by default (details)
290. configure/nvml: move some code inside a conditional (details)
291. configure: cleanup pthread_mutex_lock checks (details)
292. configure: add some messages between backend-specific checks (details)
293. configure/rsmi: allow to specify the ROCm install path (details)
294. utils/gather-cpuid: bring back pconfig (details)
295. contrib/android: store the gradlew +x permission in GIT (details)
296. contrib: cleanup/factorize scripts fot checking windows/android versions (details)
297. android: don't ignore the greek (details)
298. linux: use MPOL_PREFERRED_MANY when available (details)
299. linux: expose cache os_index when available (details)
300. x86: add a comment to remember why we don't expose cache os_index (details)
301. hwloc.h: clarify that get_*bind function fill the caller-given bitmap (details)
302. lstopo/draw: use a bitmask of backend flags instead of the no_half_lines field (details)
303. lstopo/draw: add flag for backends returning approximative textwidth (details)
304. lstopo/draw: don't always center-align factorized boxes and text (details)
305. lstopo/draw: no need to save this (details)
306. lstopo/draw: cleanup the management of memory-above children order (details)
307. lstopo/draw: rename --children-order memoryabove into memory:above (details)
308. lstopo/draw: some cleanups in the placement code (details)
309. lstopo/draw: place I/O and Misc separately from CPU by default (details)
310. lstopo/draw: allow to specify horiz/vert/rect layout for right/below io/misc (details)
311. lstopo.1: improvement in the LAYOUT section (details)
312. doxy: fix a typo (details)
313. rsmi: add a "XGMIHops" matrix (details)
314. rsmi: add missing error test in previous commit (details)
315. configure: use ROCm config for detecting OpenCL (details)
316. hwloc-calc.1: add an example for resctrl bitmap strings (details)
317. NEWS: cite the XGMIHops matrix (details)
318. linux: add "bxi0" OFED OS devices for Atos/Bull BXI HCAs (details)
319. core: fix a corner case of reconnecting levels during the merging of identical level (details)
320. core: rename hwloc_connect_io_misc_levels() (details)
321. tests/linux: update outputs (details)
322. tests/linux: cleanup fake PCI sysfs files for nvidia GPUs with NUMA nodes (details)
323. pci: consider unclassified devices as important (details)
324. gather-cpuid: update to AMD 0x80000021 leaf (details)
325. x86: AMD Family 19h (Zen3) only reports 1 node per proc (details)
326. x86: fix a comment in AMD quirk (details)
327. x86: update the AMD L3 APIC ID sharing quirk for Family 19h (Zen3) (details)
328. hwloc-annotate: typo in --help (details)
329. doxy: improve the FAQ about asymmetric topologies (details)
330. hwloc-annotate: add subtype command for setting/changing object subtype (details)
331. hwloc-annotate: add a command to change cache/memory sizes (details)
332. utils: add a test of building a custom heterogeneous topology (details)
333. doxy: add a FAQ about creating a fake asymmetric/heterogeneous topology (details)
334. utils: remove version infos from test-build-custom-topology (details)
335. rsmi: add VRAM and GTT memory info attributes (details)
336. doxy: add a FAQ entry about ROCm SMI configure options (details)
337. doxy: fix DDR/DRAM mix in FAQ about build a custom topo (details)
338. doxy+test-build-custom-topology: use the modern annotate command-line layout (details)
339. VERSION: master is now 2.7 (details)
340. doxy: L0 OS devices are of subtype COPROC, not GPU (details)
341. dolib: don't use _snprintf when snprintf is available (details)
342. core: fix a #if vs #ifdef for constructor attribute support (details)
343. core: -no-undefined is now supported on Windows (details)
344. private/misc.h: use HWLOC_HAVE_ATTRIBUTE_UNUSED instead of __HWLOC_HAVE_ATTRIBUTE_UNUSED (details)
345. cpukinds: don't duplicate an empty array of cpukinds (details)
346. contrib: add CMakeLists.txt under contrib/windows-cmake (details)
347. git-hooks: check Windows-CMake version (details)
348. contrib/ci.inria.fr: add a Windows-CMake job (details)
349. doxy: improve the FAQ about building for Windows (details)
350. .gitignore: update for test-build-custom-topology (details)
351. configure.ac: fix copyright prefix character (details)
352. build/windows: directly call MSLIB instead of going through our dolib (details)
353. build: remove dolib.c (details)
354. configure: remove CC_FOR_BUILD (details)
355. contrib/android: cleanup cmake requirement (details)
356. contrib/android: cleanup version code management (details)
357. ci.inria.fr: check the cmake version too (details)
358. git-hooks: add --update to android/check-versions.sh (details)
359. git-hooks: add --update to windows/check-versions.sh (details)
360. git-hooks: add --update to windows-cmake/check-versions.sh (details)
361. contrib: add update-versions.sh (details)
362. contrib/android: update/fix gradle-wrapper (details)
363. utils: fix a MSVC warning about size_t/int casts (details)
364. lstopo: fix some MSVC warnings about implicit casts (details)
365. xml: fix a MSVC warning about cast (details)
366. cpukinds: fix the sorting of cpukinds by ranking_value (details)
367. xml/export/v1: fix distances export vs v1 memory group insert (details)
368. API: clarify the documentation of hwloc_alloc_membind_policy() (details)
369. doxy: use &lt;version&gt; instead of <version> (details)
370. doxy: add "(FAQ)" after "Frequently Asked Questions" title (details)
371. doxy: fix ROCm build anchor and add a crossref (details)
372. doxygen: remove some obsolete config options to fix warnings (details)
373. doxygen: a4wide support has been dropped (details)
374. linux: don't look at packages if we're not the first PU of the core (details)
375. linux: add "Cluster" groups for Linux 5.16+ cluster topology (details)
376. linux: add HWLOC_DONT_MERGE_CLUSTER_GROUPS envvar (details)
377. tests/linux: add a ARM kunpeng 920 with a cluster level between L2 and L3 (details)
378. windows: improve debug printf about processor groups (details)
379. windows: move some common definitions to private/windows.h (details)
380. windows: cleanup the management of windows query structures (details)
381. windows: support multiple processor Groups per NUMA (details)
382. windows: support multiple processor Groups per cache (details)
383. windows-cmake: publicize user-facing options (details)
384. windows-cmake: Configure dynamically based on platform capabilities (details)
385. windows-cmake: use setter methods, msvc-only options to allow non-msvc (details)
386. windows-cmake: use modern install() method syntax (details)
387. windows-cmake: hwloc object for deduplicated internal logic (details)
388. windows-cmake: add tests (details)
389. windows-cmake:CI: use CMake CTest (details)
390. windows-cmake: add WIN32 option for GUI no-console (details)
391. windows-cmake: search for LibXml2 (details)
392. windows-cmake: auto-configure static-components.h (details)
393. windows-cmake: optional cuda enabled (details)
394. windows-cmake: conditional enable x86 component (details)
395. windows-cmake: do not force lib filename prefix on MSVC (details)
396. windows-cmake: make explicit options for libxml2,cuda,opencl (details)
397. windows-cmake: better BUILD_SHARED_LIBS default (details)
398. doxy: list all subtypes and add cross-refs in the obj attr section (details)
399. hwloc-calc.1: improve the resctrl bitmask conversion (details)
400. lstopo.1: fix some missing/misplaced closing \fB...\fR tags (details)
401. lstopo: properly skip --top-color argument after processing (details)
402. lstopo/draw: cleanup default object colors (details)
403. lstopo/draw: cleanup line color management (details)
404. lstopo/draw: rename colors into color_list (details)
405. lstopo/draw: don't use the "misc" color for "group" (details)
406. lstopo/draw: define a "palette" structure for all colors (details)
407. lstopo/draw: add --palette grey for greyscale graphical output (details)
408. lstopo/draw: add a white palette to disable all background colors (details)
409. lstopo/draw: allow #xxyyzz in --binding/disallowed-color (details)
410. lstopo/draw: cleanup --top-color (details)
411. lstopo/draw: add --palette type=#rrggbb to tune colors (details)
412. windows: don't create groups for windows processor groups by default (details)
413. NEWS: clarify the NEWS entry about the new windows API (details)
414. NEWS: move Windows entries together (details)
415. windows: remove a unneeded check (details)
416. windows: fix the source string when inserting processor group objects (details)
417. windows: remove support for the old non-"Ex" GetLogicalProcessorInformation() (details)
418. lstopo/windows: fix the initial window size by using the right fonts (details)
419. contrib/windows-cmake: udpate private_config.h.in (details)
420. NEWS: add bullet about many Windows CMake improvements (details)
421. lstopo/draw: rename our RGB macros into LSTOPO_COLOR (details)
422. levelzero: don't skip devices if sysman is missing (details)
423. levelzero: only warn once if sysman is disabled (details)
424. levelzero: update the management of vendor/device/brand/... strings (details)
425. levelzero: add info attrs about type, nr slices, subslices, EUs and threads (details)
426. levelzero: factorize the querying of CQ groups (details)
427. levelzero: expose subdevices as sub-osdevices (details)
428. level: fix the querying of subdevice properties (details)
429. levelzero: fix the test for unknown modelname before adding info attrs (details)
430. windows: update comments about new topology features in recent versions (details)
431. levelzero: update some comments (details)
432. levelzero: query memory and expose LevelZeroHBM/DDR/MemorySize info attrs (details)
433. levelzero: remove Device in some info attr name (details)
434. lstopo/draw: add some L0 attributes in osdevs (details)
435. doxy: fix some typos in new LevelZero info attrs (details)
436. doxy: small cleanups in LevelZero attrs (details)
437. linux/cpukinds: ignore max frequencies is base frequency is known (details)
438. NEWS cleanup before branching 2.7 (details)
439. VERSION: master is now 2.8 (details)
440. autogen: exit on first error (details)
441. autogen: hide bigsur libtool patch apply failure messages (details)
442. contrib/android: google play wants targetSdkVersion 30+ (details)
443. linux/cpukinds: adjust max frequencies instead of ignoring them (details)
444. levelzero: ignore DDR memory if the device is integrated (details)
445. contrib/android: bump compileSdkVersion to 30 as well (details)
446. xml: support id="objXYZ" instead of gp_index="XYZ" (details)
447. README: irc channel moved from Freenode to Libera Chat (details)
448. doxy: also fix the IRC server here (details)
449. darwin: kIOMasterPortDefault is deprecated (details)
450. utils/common-ps.c: reduce a variable scope (details)
451. configure: fix --with-hwloc-symbol-prefix help string (details)
452. core: realign the long list of allowed topology flags (details)
453. API: add topology flag NO_DISTANCES (details)
454. API: add topology flag NO_MEMATTRS (details)
455. API: add topology flag NO_CPUKINDS (details)
456. doxy+README: talk about new flags for ignoring distances, memattrs and cpukinds (details)
457. doxy: improve the doc about filtering objects to reduce the overhead (details)
458. API: minor clarification: topology flags are 0 by default (details)
459. contrib/ci.inria.fr: never disable make check in coverity and sonar-scanner jobs (details)
460. darwin: factorize a range of consecutive bitmap_set() (details)
461. darwin: split to a generic cpukinds query and register (details)
462. darwin: abstract out the building of numa and cache levels (details)
463. darwin: optionally get cpukinds from new sysctl hw.perflevels too (details)
464. darwin: detect heterogeneous caches from sysctl perflevels (details)
465. contrib/windows-cmake: install hwloc.pdbg with Debug shared libs (details)
466. doxy: fix a typo in L0 attr description (details)
467. levelzero: improve the error message on ze_init() failure (details)
468. levelzero: verbose messages about number of drivers and devices (details)
469. levelzero: add LevelZeroUUID (details)
470. core+levelzero: Set ZES_ENABLE_SYSMAN via setenv instead of putenv (details)
471. darwin: read cachelinesize before creating hybrid perflevel caches (details)
472. core: fix setenv(ZES_ENABLE_SYSMAN) on mingw (details)
473. core+levelzero: fix typos in ZES_ENABLE_SYSMAN enabling (details)
474. doc: Fix typos in hwloc.h (details)
475. Strengthen the NVML check by adding a runtime check (details)
476. doc: fix typos + harmonize parameter name (details)
477. configure: add HWLOC_FOO_COMPONENT_BUILTIN macros to public headers (details)
478. configure: only show the NVML status line if we have the big warning (details)
479. configure: move plugin-path AC_ARG_WITH with others (details)
480. configure+doxy: document that --with-cuda=/nonexisting will disable everything installed by CUDA (details)
481. configure: also warn if rsmi lib isn't in LD_LIBRARY_PATH and ld.so.conf (details)
482. configure: also warn if cudart lib isn't in LD_LIBRARY_PATH and ld.so.conf (details)
483. doxy: object attributes for sizes are in KiB instead of kB (details)
484. synthetic: accept memory/cache size as GiB/KiB and handle GB/kB correctly (details)
485. utils/annotate: accept memory/cache size as GiB/KiB and handle GB/kB correctly (details)
486. tests: update to synthetic/annotate now handling GB/GiB correctly (details)
487. lstopo.1: add a sentence about kB/KiB units (details)
488. API: bump HWLOC_API_VERSION to 0x20800 (details)
489. API+lstopo.1: clarify that disallowed doesn't mean offline (details)
490. core: don't enable L0 Sysman if L0 is disabled (details)
491. memattrs.h: use a sentinel MAX value for memattr IDs (details)
492. memattrs: add Read/Write Bandwidth/Latency attributes (details)
493. linux: gather read/write bandwidth/latency (details)
494. lstopo/memattrs: abstract-out a function to display a single memattr (details)
495. lstopo/memattrs: reorder memattrs in the output (details)
496. tests/x86: add a Zen3 testcase (details)
497. x86: workaround VMs reporting invalid core/thread info (details)
498. NEWS: import 2.7.1 bullets in master (details)
499. configure: add --disable-readme (details)
500. lstopo.1: clarify the documentation about PCI link speed in the graphical output (details)
501. utils: allow loading fsroot/cpuid topologies from a tarball (details)
502. tests/lstopo: test archivemount of fsroot and cpuid (details)
503. distances: fix clang warning about NULL pointer substraction in container_offset (details)
504. utils/gather-cpuid: update comments about cpuid leaves reserved on Intel/AMD (details)
505. utils/gather-cpuid: update with Intel CPUID leaves 0x1c and 0x21 (details)
506. utils/gather-cpuid: update for AMD CPUID leaf 0x80000022 (details)
507. utils/gather-cpuid: disable cpukinds/memattrs/distances in the topology (details)
508. lstopo: add --os/logical-index-prefix (details)
509. levelzero: use the core PCI prop extension if available (details)
510. pci: rework locality quirks (details)
511. pci: reduce the scope of pci-discovery-prepared phases (details)
512. pci: add a locality quirk for the HPE Cray EX235a (details)
513. pci: add a fake pci locality quirk for testing (details)
514. NEWS: improve Linux cpukinds max freq item (details)
515. linux/io: move block sectorsize related lines together (details)
516. linux/io: remove support for /sys/class/dax (details)
517. linux/io: cleanup dax/non-dax devtype management (details)
518. linux/io: no need for sysfs local_cpus for OSdev locality (details)
519. linux/io: rework/fix numa_node attribute in sysfs (details)
520. gather-topology: gather dax driver info (details)
521. tests/linux: add dax driver information to gathered files (details)
522. linux: fix and factorize the checking of whether a DAX device is exposed as NUMA node (details)
523. linux: add DAXParent and DAXType info attr (details)
524. linux/dax: add some comments (details)
525. linux/block: replace "NVDIMM" subtype with "NVM" or "SPM" to match DAX attributes (details)
526. memattrs: heuristics to set NUMA node subtype to DRAM/HBM/SPM/NVM (details)
527. tests: add memtiers for testing subtypes of heterogeneous memory nodes (details)
528. tests/linux: add a complex test case with lots of heterogeneous memories (details)
529. NEWS: some improvements in 2.8 bullets (details)
530. pci: Gen7 bandwidth almost confirmed to be twice of Gen6 (details)
531. cuda: update to compute capability 8.6 (details)
532. pci: also update the computation of Gen6+ linkspeed (details)
533. private/windows.h: guard against _ANONYMOUS_UNION/STRUCT redefinition (details)
534. VERSION: master is now 2.9 (details)
535. tests/linux: remove 1alpha (details)
536. tests/linux: cleanup missing cpus in nvidiagpunumanodes (details)
537. linux: improve the readdir'ing of /sys/devices/system/{cpu,node} (details)
538. linux: abort if sysfs cpu topology is missing (details)
539. linux: don't support both /sys/bus/cpu/devices and /sys/devices/system/cpu (details)
540. linux: stop using /sys/bus/node/devices too (details)
541. linux: don't check whether numa topology is properly available in sysfs (details)
542. linux: simplify the checks for sysfs cpu file availibility (details)
543. utils/calc: don't segfault if IO is given to -N or -I (details)
544. utils/calc: allow -H with NUMA nodes (details)
545. pci: document the PCI affinity quirk for HPE Cray EX235a (details)
546. utils/tests/calc: check for multiple levels of groups (details)
547. utils/calc: allow to retrieve object attributes when scanning type strings (details)
548. utils/calc: cleanup the management of -I (details)
549. utils/calc: support osdev subtypes in -N and -I (details)
550. levelzero: fix subdevice info properties (details)
551. Add codemeta file (details)
552. pci+linux: make 32bits domain warnings non-critical (details)
553. core: if debug is enabled and debug messages are disabled, unhide errors too (details)
554. xml: also demote 32bits pci domain import errors to non-critical (details)
555. core: add HWLOC_SHOW_ALL/CRITICAL_ERRORS() macros (details)
556. xml: fix an error message in the future xml obj id attribute (details)
557. completion: fix diff --refname (details)
558. Add more codemeta information (details)
559. bitmap.h: fix the doc for foreach() (details)
560. doc/examples: add memory-attributes.c (details)
561. memattrs.h: cleanup doxygen formatting of the enum of memory attributes (details)
562. doc/examples: use hwloc_uint64_t in memory-attributes.c (details)
563. nvml/bw: add nvlink versions 3&4 (details)
564. tests/backends: fix build with ClangCI 15 compiler frontend for MSVC (details)
565. completion/lstopo: update the list of output formats (details)
566. lstopo: if stdout is redirected, use console mode (details)
567. lstopo: only show interactive help if stdout is a tty (details)
568. lstopo: add an explicit "window" output format (details)
569. cuda: update to compute capability 9 (details)
570. Leak in hwloc__xml_import_cpukind (details)
571. lstopo/draw: fix the displaying of non-HBM L0 device memory (details)
572. cuda: update for compute capability 8.9 (details)
573. doxy: number of character +s (details)
574. diff: reuse temp variable (details)
575. pci: don't ignore CXL memory expanders (details)
576. utils/gather-topology: gather cxl devices and drivers (details)
577. levelzero: build the "XeLinkBandwidth" distance matrix (details)
578. NEWS: more bullets for 2.9 (details)
579. linux+lstopo/draw: expose CXL memory device (Type 3) size (details)
580. VERSION: master is now 2.10 (details)
581. levelzero: save some disabled code to get driver properties (details)
582. VERSION: master is now 3.0 (details)
583. API: always enable 32bits PCI domains (details)
584. pci: add the prog-if (register level programming interface) to pci device attributes (details)
585. rename.h: replace hwloc_ with HWLOC_ when renaming capital names (details)
586. BGQ: remove support (details)
587. core: merge is_loaded/is_thissystem ints into a bitwise state (details)
588. core: add is_init and is_loading topology states (details)
589. deprecated.h: remove functions deprecated since 2.0 (details)
590. pci: remove deprecated HWLOC_PCI_<domain>_<bus>_LOCALCPUS envvars (details)
591. helper: remove an obsolete ref to an old deprecated-removed function (details)
592. API: add hwloc_get_type_depth_with_attr() (details)
593. API: deprecate hwloc_type_sscanf_as_depth() (details)
594. API: replace hwloc_obj_type/attr_snprintf() verbose attribute with flags (details)
595. misc: replace hwloc_memory_size_printf_unit/value() with hwloc_memory_size_snprintf() (details)
596. obj_attr_snprintf: switch to KiB/MiB/GiB/TiB (details)
597. lstopo/draw: switch CUDA MP local memory to KiB too (details)
598. obj_snprintf(): add flags to tune type names and units (details)
599. lstopo: add --osf/--obj-snprintf-flags option to tune obj printing (details)
600. VERSION: update to final 2.9.0 bullets (details)
601. hwloc-gather-topology: only unalias ls (details)
602. contrib/android: bump compile/targetSdkVersion to 31 (details)
603. contrib/android: update build parameters for Android 12 (details)
604. contrib/android: bump android app subversion to 1.5.2 (details)
605. NEWS: bring 2.7.2 bullets (details)
606. contrib/*/check-versions.sh: add --update to usage (details)
607. contrib/android/check-versions.sh: make a remaining verbose message conditional (details)
608. contrib/dist/make_dist_tarball: update android/windows VERSION too (details)
609. doxy: example using get_api_version() to manage flags at runtime (details)
610. test-gather-topology: skip if /sys isn't available (details)
611. utils/gather-cpuid: update to Intel x86 Manual 2022/12 (details)
612. utils/gather-cpuid: always stop ecx loops after 256 iterations (details)
613. x86: limit the number of cache and topology subleaves (details)
614. lstopo: improve the error message in case of unrecognized option (details)
615. lstopo.1: typo (details)
616. lstopo: remove .format from -.format special output early (details)
617. lstopo: show the output format in some error messages (details)
618. lstopo: update the output format when it is internally chosen (details)
619. lstopo: warn that filename is ignored when forcing the window output (details)
620. lstopo: show a message about the export filename and format (details)
621. API/deprecated: remove object type backward compat macros for v1.x (details)
622. hwloc-calc: remove support for obsolete options --pulist --nodelist and --objects (details)
623. tools: remove support for obsolete --xml/--fsys-root/--synthetic options (details)
624. core: don't stop parsing the "socket" name for packages (details)
625. tests: filter-out the upcoming "id" xml attribute (details)
626. xml: add a v2 export flag and enable it by default for now (details)
627. xml: add support for 3.0 xml format (details)
628. xml: export id="objXXX" in XML 3.0 (details)
629. xml: stop forcing v2 export by default (details)
630. doc/images: update some xml source to XML v2 (details)
631. doc/images: regenerate lstopo outputs (details)
632. tests/xml: remove tests importing from v1 (details)
633. tests/netloc: update XMLs to v2 (details)
634. xml: drop support for importing XML v1 (details)
635. tests/lstopo: check v2 XML export instead of v1 (details)
636. tests/xml: remove tests exporting to v1 (details)
637. xml: remove support for exporting to XML v1 (details)
638. dtd: remove hwloc.dtd now that it's not supported anymore (details)
639. xml/import: remove the dead code (only used for v1 import) (details)
640. tests/xml: remove v1tov2 options files (details)
641. darwin: don't forget to filter perflevel caches (details)
642. API: rename Block OS devices into Storage (details)
643. API: split a "Memory" osdev type out of "Storage" (details)
644. xml: import/export memory osdev from/to xml 2.0 (details)
645. tests/xml: check import/export from/to XML v2 (details)
646. linux: force an ending \0 in our readlink (details)
647. linux/dax: don't reuse block-specific sysfs parsing code (details)
648. linux/block: simplify the code now that DAX support is separated (details)
649. tests/backends.c: fix a failure when /sys isn't available on Linux on non-x86 (details)
650. hwloc-gather-cpuid: update some comments (details)
651. hwloc-gather-cpuid: add 0x80000023 (details)
652. hwloc-gather-cpuid: add AMD 0x80000026 extended topology leaf (details)
653. .github/issue_template.md: improve the list of command outputs requested from non-Linux OS (details)
654. linux: mark HPE Cray Cassini HSNs with subtype "Slingshot" (details)
655. configure: update to latest ROCm include/lib paths (details)
656. tests/xml: rewrite the framework (details)
657. tests/xml: remove unneeded environment variables (details)
658. tests/xml: remove 8intel64-fakeKNL-A2A-hybrid.rootattachednumas (details)
659. tests/{x86,linux,xml...}: uniformize the path to LOG_COMPILER (details)
660. tests/linux*: look for .source and .exclude in the directory of .output (details)
661. Update .gitignore to include cross-compile libtool artifacts (details)
662. tests/xml: some renaming (details)
663. tests/xml: improve the format of .test files (details)
664. tests/linux: remove some of the many tests that aren't useful anymore (details)
665. tests/linux: rewrite the framework with .test files (details)
666. tests/linux/allowed: rewrite the framework with .test files (details)
667. tests/x86+linux: rewrite the framework with .test files (details)
668. tests/x86: rewrite the framework with .test files (details)
669. nvml: fix nvlink 3/4 bandwidth in our distances matrix (details)
670. restrict: fix when doing nothing except because of REMOVE_CPULESS (details)
671. tests/restrict: don't make the topology a global variable (details)
672. tests/restrict: add some redundant restrict() (details)
673. tests/{linux,xml,x86,x86+linux,linux/allowed}: allow to run tests in an optional extra subdirectory (details)
674. NEWS: bring 2.9.1rc1 bullets in master (details)
675. synthetic/export/memory: replace a buggy assertion with some warnings (details)
676. synthetic: typo in error message (details)
677. utils/archivemount: mount read-only (details)
678. utils: don't require archivemount at configure-time (details)
679. utils/archivemount: disable on !linux (details)
680. core: fix group->total_memory after hwloc_topology_insert_group_object() (details)
681. synthetic: factorize/fix default attributes (details)
682. core: fix the updating of total_memory in hwloc_topology_insert_group_object() (details)
683. synthetic/import: ignore unknown attributes instead of failing (details)
684. synthetic/export/v1: ignore memcaches instead of failing to export (details)
685. synthetic/import: allow memory-side caches (details)
686. synthetic/export: export memory-side caches (details)
687. Add security policy (details)
688. hwloc_obj_add_other_obj_sets: handle bitmap realloc errors (details)
689. hwloc/glibc-sched.h: handle bitmap realloc errors (details)
690. hwloc/linux-libnuma.h: handle bitmap realloc errors (details)
691. hwloc.h: many small improvements about return values and errors (details)
692. hwloc/shmem.h: add/improve doc about return values (details)
693. IO helpers: clarify that some IO helpers return 0 on success and -1 on error (details)
694. hwloc/linux.h: document 0 on success and -1 on error (details)
695. hwloc/memattrs.h: improve doc about return values (details)
696. hwloc/export.h: improve doc about return values (details)
697. hwloc/diff.h: improve the documentation about return values (details)
698. hwloc/distances.h: add/improve doc about return values (details)
699. hwloc/helpers.h: document return values of many helpers (details)
700. hwloc.h: rewrite the doc of hwloc_get_type_depth() (details)
701. hwloc.h: doc return values for two remaining functions (details)
702. hwloc/bitmap.h: improve doc about return values (details)
703. API: update one very old comment about v1 multi-node topology (details)
704. doxy: remove broken link to XML v1 export flag (details)
705. hwloc/helpers.h: fix a type (details)
706. hwloc.h: add an error reporting section describing the common conventions (details)
707. utils/archivemount: propagate error when visiting subdirs (details)
708. x86: document the array of cpu features (details)
709. x86: use a global is_hybrid flag (details)
710. x86: remove the explicit apic_number in leaves 0xb/0x0f (details)
711. gather-cpuid: note about future AMD 0x8000002[78] leaves (details)
712. x86: Intel has a new "DieGrp" level in the CPUID doc (details)
713. x86: only try to build die objects and unit/tile/module groups if we found some info (details)
714. x86: update the loop break when reading topology cpuid leaves 0x1f/0x0b (details)
715. x86: add support for AMD 0x80000026 CPUID leaf (details)
716. tests/x86: add a amd zen4 genoa case for cpuid leaf 0x80000026 (details)
717. linux: detect fake numa in the Linux kernel cmdline (details)
718. linux/numa: cleanup HWLOC_DEBUG_ALLOW_OVERLAPPING_NODE_CPUSETS (details)
719. linux/numa: disable memory attributes and memory-side cache in fake numa (details)
720. lstopo/draw: cleanup horiz/vert layout management (details)
721. lstopo/draw: make memory:above layout configurable and use rectangular by default (details)
722. Fix name scheme when building with --target (details)
723. linux/dax: cleanups and comments in annotate_dax_parent() (details)
724. linux/dax: add CXLDevice info attribute(s) in DAX and NUMA nodes (details)
725. xml/import: allow (and ignore for now) info children in more xml tags (details)
726. xml/import: fix a leak when ignoring distances infos (details)
727. doxy: non-sysfs Linux support was removed (details)
728. doxygen: update the config to 1.9.4 (details)
729. doxy: ref the website language bindings instead of duplicating it here (details)
730. Fix typos in hwloc README and netloc Doxygen. (details)
731. doxy: update doxy from README (details)
732. netloc_ib_extract_dats.c: closedir() the right dir on error (details)
733. utils/archivemount: Restore working directory (details)
734. netloc: remove (details)
735. doxy: merge main.doxy back into hwloc.doxy (details)
736. README: update the build after netloc removal (details)
737. lstopo.1: xml support is always available (details)
738. lstopo: allow output format/extension v2xml/v3xml (details)
739. linux/cpukinds: use Intel PMU directories to detect Atom/Core cpusets on Intel hybrid (details)
740. gather-topology: gather Intel atom/core pmu "cpus" files (details)
741. doxy: typo (details)
742. x86: enable CPUID 0x1f leaf (intel v2 topo enumeration) on Zhaoxin (details)
743. hwloc-calc.1: fix the formatting of --cpukinds (details)
744. hwloc-calc: improve --cpukind parsing (details)
745. hwloc-calc/bind.1: more info about --no-smt (details)
746. hwloc-calc/bind.1: more info about --cpukind (details)
747. x86: fix/cleanup the breaking-out of the topology enumeration loops (details)
748. levelzero: typo in some error messages (details)
749. fix prefix before error messages (details)
750. levelzero: typo in error message (details)
751. levelzero: don't initialize anything before checking if I/O is enabled (details)
752. pci: don't initialize anything before checking if I/O is enabled (details)
753. pci: factorize multiple calls to filters (details)
754. doxy: split out a faq about disabling things (details)
755. doxy: improve the faq entry about slow gpu discovery (details)
756. doxy: improve the text about keep structure in the new "disable" faq entry (details)
757. doxy: typo (details)
758. levelzero: fix a snprintf overflow warning (details)
759. utils/annotate: fix [kMGT]B size units (details)
760. utils/hwloc: check topology loading error status (details)
761. utils/hwloc-ps: Fix output stream when using '-h' (details)
762. linux: add hardwired fallback for syscall numbers on riscv (details)
763. utils/hwloc: enforce using '\0' instead of 0 (details)
764. core+lstopo: don't forget to set the L3i filter when operating on all caches (details)
765. core: remove some redundant checks when setting filters (details)
766. core: allow backends to check if all cpu caches are filtered-out (details)
767. doxy: don't mix components and plugins in the text (details)
768. doxy: try to clarify terms component/backend/plugin (details)
769. doxy: don't duplicate minimalistic stuff about xml in the component page (details)
770. doxy: random improvements in the component page (details)
771. doxy: talk about disabling components in the make hwloc faster FAQ entry (details)
772. levelzero: use zesInit() when available (details)
773. tests/ports: add some missing -D to enable more ports' code (details)
774. contrib/dist/publish_doc: update the example doc we look for (details)
775. contrib/dist/publish_doc: symlink to the latest common jquery.js (details)
776. windows: Fix type of hostname_size parameter (details)
777. darwin: fix IOKit warnings about const char * being passed in arrays (details)
778. contrib/dist/pubish_doc: further improve jquery.js (details)
779. NEWS: bring 2.9.2 bullets in master (details)
780. hwloc.h+bind.1in: add a note about hwloc bind policy != linux bind (details)
781. doxy: improve the cpu_mem_bind chapter (details)
782. doxy: move the installation section to its own page (details)
783. doxy: improve the main page (details)
784. README: update after doxy main page reorganization (details)
785. linux/cxl: merge CXLDevice attributes when interleaved (details)
786. doxy: remove our custom doxygen.css for html docs (details)
787. doxy: move bugs/questions to the end of the introduction page (details)
788. components: allocate the backend private_data together with the backend structure (details)
789. cuda/levelzero/gl/nvml/opencl/rsmi: add a single Backend info attr in the root (details)
790. hardwired: add a Backend info attr (details)
791. noos: add a Backend info attr (details)
792. pci: add a Backend info attr (details)
793. freebsd: make sure uname and Backend info attrs are added once (details)
794. linux: make sure uname, Backend and other global info attrs are added once (details)
795. xml/export: factorize info attr export (details)
796. xml/export/v2: add Backend info back to OS devices (details)
797. xml/import/v2: add Backend info to the root object from OS devices (details)
798. xml/import: cleanup unused/non-unused function parameters (details)
799. diff: also check whether cpukinds are identical (details)
800. XML: fix a typo in comments (details)
801. xml: remove the nbnumanode import temporary data (details)
802. API: the non-printable character warning applies to info key names too (details)
803. doc: clarify how we talk about info attributes (details)
804. doxy: add missing comma (details)
805. API: add hwloc_infos_s structure in objects (details)
806. core: switch internal functions to use hwloc_infos_s structure (details)
807. API: add hwloc_get_info_by_name() (details)
808. cpukind: switch to struct hwloc_infos_s in the API (details)
809. core: add an "allocated" private field in struct hwloc_infos_s (details)
810. API: make it clear that add_info() wants non-NULL name+value (details)
811. API: add hwloc_modify_infos() to add/replace/remove info attrs (details)
812. tests: rename and enlarge the infos test (details)
813. core: use the new internal hwloc__replace_infos() and remove add_info_nodup() (details)
814. utils/annotate: abstract-out info attr operations (details)
815. utils/annotate: reimplement info operations using hwloc_modify_infos() (details)
816. utils/annotate: add the "cpukind#N" fake location to tweak cpukind infos (details)
817. tests/{cuda*,nvml,opencl,levelzero,rsmi,gl}: fix Backend info attr check (details)
818. utils/annotate.1: remove a spurious '+' (details)
819. rsmi: remove some warnings on 32bits (details)
820. linux: fix a warning about fgets return value not being used. (details)
821. utils/ps: fix a warning about read() return value not being used (details)
822. utils/misc/archivemount: fix warnings about unchecked return values of chdir and system (details)
823. components: fix a warning about unused asprintf return value (details)
824. darwin: fix some printf format warnings (details)
825. API: add topology-specific info attributes (details)
826. utils/annotate: "topology" location may be used to manipulate topo attributes (details)
827. utils/info: --topology -v shows topo info attrs (details)
828. lstopo/text: show topo info attrs in verbose mode (details)
829. core: move some root info attrs to topology infos (details)
830. xml/import/v2: move v2 root info attrs to topo when appropriate (details)
831. xml/export/v2: export topo info attrs as root info in v2 (details)
832. contrib/windows-cmake: move tests/hwloc/CMakeLists.txt back to ./tests/ (details)
833. shmem: duplicate topo infos so that we may modify them (details)
834. allow: now allowed on shmem-adopted topologies (details)
835. API: add notes about restrict/insert_group changing logical indexes (details)
836. API: clarify that restrict() removes PUs or NUMAs (details)
837. utils: clarify that --restrict removes PUs and/or NUMA nodes (details)
838. utils: clarify that --restrict may cause logical index to change (details)
839. API: return meaningful positive values from modify_infos() (details)
840. utils/ps: add --children-of-pid (details)
841. lstopo: abstract more the code to insert misc children (details)
842. lstopo: add --misc-from (details)
843. hwloc-ps: add --lstopo-misc to output something for --lstopo --misc-from (details)
844. doxy: remove way too many spaces (details)
845. x86: add a debug messages when disabling because of non-unique APIC ids (details)
846. core: don't crash when clearing an already empty topology (details)
847. linux/smbios: enforce memory header from spec 2.3 only (details)
848. linux: add CXL device serial number (details)
849. linux: add more MemoryModule info attributes (details)
850. xml/import: remove "v2" from some function names (details)
851. xml/import/v2: check version major before enabling some info checks (details)
852. bitmap.h: clarify that the empty bitmap doesn't intersect anything else (details)
853. utils/calc: fix searching CPU objects below heterogeneous NUMA (details)
854. utils/calc: fix searching NUMA nodes inside CPU objects when NUMA are attached higher (details)
855. android: disable the platform database for now (details)
856. contrib/android: bump compile/targetSdkVersion to 33 (details)
857. utils/calc: add --object-output (details)
858. utils/bind: warn if neither CPU nor memory binding location were given (details)
859. utils/bind: always warn when assuming the executable starts at an unrecognized argument (details)
860. utils/calc.1: fix the doc of --largest (details)
861. helpers.h: fix missing MemCache in get_next_child() (details)
862. helper.h: move type/kind section above others (details)
863. helpers.h: cleanup get_next_child() (details)
864. utils: add -q and --quiet when we have -s and --silent (details)
865. linux: handle glibc cpuset allocation failures (details)
866. tests/linux: add a testcase with memory side caches (details)
867. tests/xml: add a testcase with memory side caches (details)
868. utils/info: factorize the displaying of ancestors (details)
869. utils/info: factorize the displaying of descendants (details)
870. utils/info: abstract-out displaying of children, local memory and single objects too (details)
871. utils/info: add [kind=]cpu/cache/normal/memory/io/all to --ancestor and --descendants (details)
872. utils/info: add --first to stop on first matching object (details)
873. doxy: add "Object Kind" in terms and defs (details)
874. lstopo: fix --osf in usage (details)
875. doxy: fix a crossref to hwloc_modify_infos_op_e (details)
876. doxy: fix a crossref to hwloc_disc_status_flag_e (details)
877. contrib/ci.inria.fr: use UTC time for snapshot name (details)
878. contrib/android: fix missing text outside of boxes (details)
879. contrib/android: rework the drawing of lines (details)
880. contrib/android: disambiguate multiple box IDs to fix the text in collapsed PCI devices (details)
881. contrib/android: also disable PCI collapsing when we disable factorization (details)
882. contrib/android: bump app to 1.5.3 (details)
883. doxy: typo (details)
884. tests/hwloc_type_sscanf: check osdev specific type too (details)
885. obj_type_snprintf: show "OS[type]" or "OSDev[type]" instead of "type" by default (details)
886. tests/type_sscanf: update for new osdev format (details)
887. xml/import/v2: update the osdevice type before filtering (details)
888. osdev: change osdev.type from an enum to a bitmask (details)
889. osdev: BXI is now Network instead of OpenFabrics (details)
890. osdev: use multiple bits in the osdev-specific type (details)
891. tests: update test-hwloc-compress-dir XMLs to new OS device types (details)
892. gather-topology: bring back /proc/device-tree/cpus if --dt is given (details)
893. NEWS: bring 2.9.3 bullets in master (details)
894. core/insert_group: free the input object if failing to insert because shmem topo (details)
895. core/insert_group: free the group if merging with root (details)
896. tests/groups: check failing insert (details)
897. API: document that insert_group() frees the input object on failure to insert (details)
898. API: document that group alloc/insert must use the same topology (details)
899. API: add hwloc_topology_free_group() (details)
900. API: more links between hwloc_topology_alloc/insert/free_group_object() (details)
901. xml: fix statically-allocated opaque backend info array (details)
902. shmem: pad the end of header up to a multiple of pointer size (details)
903. doxy: update non-documented env vars (details)
904. core: improve insertion error messages (details)
905. linux: get CPUModel and CPUFamily from /proc/cpuinfo on LoongArch (details)
906. API: clarify in filter doc that MemCaches are ignored by default (details)
907. API: clarify the default filtering of dies (details)
908. core: don't let hwloc_topology_check() test uninitialized topologies (details)
909. API: clarify that XML input errors are reported either in set_xml*() or load() (details)
910. API: clarify that set_pid() assumes the PID won't change until load() (details)
911. utils: fix the filtering of osdevs by osdev.type bitmask (details)
912. utils/calc.1: fix examples with --local-memory-flags (details)
913. contrib/ci.inria.fr: shorten the build path in windows cmake (details)
914. contrib/ci.inria.fr: allow to pass local options to the Windows CMake cmdline (details)
915. utils/calc: rework parsing of levels for -I -N and -H (details)
916. utils: use the new hwloc_calc_parse_level() for parsing locations etc (details)
917. utils: use the new struct hwloc_calc_level when counting/iterating over objects (details)
918. utils: factorize the computing of the location prefix length (details)
919. utils: move the parsing of os[] and pci[:] filters to hwloc_calc_parse_level() (details)
920. utils: simplify the parsing of pci[vendor:device] filters (details)
921. utils: ignore location filters if we already parsed it as "osdev[osdev.type]" (details)
922. utils: allow to filter locations by [subtype] (details)
923. utils: factorize the filtering of objects using the new level structure with filters (details)
924. utils/tests: check location filters in calc and info tests (details)
925. utils: document location filters in calc.1 info.1 and hwloc.7 (details)
926. lstopo: rework --only using the hwloc_calc_level structure (details)
927. utils/manpages: talk about numa[hbm]:X instead of hbm:X directly (details)
928. utils/calc.h: remove a obsolete verbose message (details)
929. core: don't guess memory tiers if memattrs are disabled (details)
930. core: disable all distances/cpukinds/memattrs call if flags are set (details)
931. memattrs: rework the heuristics to find types of memory nodes (details)
932. memattrs: add "node0_is_dram" option to HWLOC_MEMTIERS_GUESS envvar (details)
933. memattrs: use a bitmask for memory tier type instead of a simple enum (details)
934. memattrs: add CXL memory tier type (details)
935. memattrs: add HWLOC_MEMTIERS envvar to force some tiers (details)
936. linux/knl: enforce DRAM subtype for KNL DDR (details)
937. memattrs: sort memory tiers by BW and expose MemoryTier info attr (details)
938. API+doxy: clarify that NUMA subtypes may be missing or wrong (details)
939. memattrs: only build memory tiers and set subtypes when not loading from XML (details)
940. utils/manpages: talk much more about HBM NUMA nodes (details)
941. utils: filter based on memory tiers using numa[tier=1] (details)
942. doxy: add a new section about heterogeneous memory (details)
943. hwloc.7: typo (details)
944. utils/gather-topology: gather "soc" bus (details)
945. linux: gather SoC information (details)
946. linux/cpukinds: separate gathering from submitting cpukind (details)
947. linux/cpukinds: adjust turboboost max frequencies before building cpukinds (details)
948. linux/cpukinds: factorize the registration of a single cpukind (details)
949. linux/cpukinds: force homogeneity on NVIDIA Grace (details)
950. API: document the linkspeed as runtime-changing (details)
951. tests/abi: check the PCI device attribute ABI (details)
952. doxy: document all PCI device/bridge attributes (details)
953. tests/x86: add a SapphireRapids Xeon test (details)
954. tests/x86: add a RaptorLake hybrid test (details)
955. tests/linux: add a RaptorLake hybrid test (details)
956. linux: fix smbios memory Misc info attrs on big endian (details)
957. linux/smbios: use endian.h htole16/32() (details)
958. test-gather-topology.sh: Disable test when cross-compiling (details)
959. hwloc/import/diff/libxml: don't count the \0 in the buffer length (details)
960. API: ending \0 must be included in xml input memory buffer and length (details)
961. xml/import/nolibxml: assume ending \0 is included in input buffer and length (details)
962. linux: fix a strncpy overflow warning (details)
963. NEWS: bring 2.10.0 bullets in master (details)
964. lstopo/graphical: only print osdev storage|memory size once (details)
965. opencl.h: use the khronos PCI bus info extension when available (details)
966. lstopo/graphics: fix the displaying of LevelZero non-HBM size (details)
967. rename: add hwloc_cl_device_pci_bus_info_khr (details)
968. contrib/windows-cmake: fix source path of cuda/opencl tests (details)
969. levelzero/xelink: correctly assign ports to subdevices (details)
970. levelzero/xelink: accumulate subdevice link bandwidths in parents (details)
971. doxy: update the description of the I/O device figure (details)
972. doxy: remove custom HTML div sections before doxygen sectons (details)
973. doxy: add a section about building with GNU Make and CMake (details)
974. [cpuid-x86.h] Add MSAN annotations to mark memory as initialized (details)
975. API: improve the documentation of insert_group() (details)
976. API: typo in previous commit (details)
977. API: add hwloc_obj_set_subtype() (details)
978. tests/hwloc/groups: use hwloc_obj_set_subtype() (details)
979. deprecated.h: remove _nodeset() functions (details)
980. doxy: update the XML compat FAQ for 3.0 (details)
981. doxy: update XML upgrade section to 3.0 (details)
982. doxy: update the API change FAQ to 3.0 (details)
983. doxy: update the API/ABI change FAQ to 3.0 (details)
984. doxy: update the 2.0 upgrade page with 3.0 (details)
985. contrib/windows-cmake: actually enable the CUDA backend when detected (details)
986. gather-topology: fix the gather of NVIDIA GPU NUMA info in /proc (details)
987. linux: don't hide the NVIDIA GPU node on non-POWER platforms (details)
988. nvml: use the current PCI link speed (details)
989. nvml: fix the PCI link speed for PCI Gen4+ (details)
990. pci+nvml: factorize PCI link speed computation (details)
991. rename.h: update with hwloc__pci_link_speed() (details)
992. info.1: typo (details)
993. windows: fix dll filenames when UNICODE is enabled (details)
994. lstopo/x11: fix Unknown icon shown on GNOME (details)
995. publish_doc: update URL checks (details)
996. doxy: don't forget to ship heteromem.png (details)
997. core: add the KiB suffix to Size info when it's implicit (details)
998. tests/xml: add "KiB" suffix to existing Size info values (details)
999. lstopo: properly handle Size info suffixes (details)
1000. doxy: document the "KiB" suffix for Size info attr values (details)
1001. utils/calc+info: rework --best-memattr to allow multiple nodes (details)
1002. utils/calc+info: abstract-out the parsing of best node flags (details)
1003. utils/bind: switch to the same code as hwloc-calc for --best-memattr (details)
1004. utils/misc: remove hwloc_utils_get_best_node_in_nodeset_by_memattr() (details)
1005. tests/x86: add a Zhaoxin KH-40000 (details)
1006. x86: update intro comments (details)
1007. x86: update comments about topology enumeration loop break-out conditions (details)
1008. tests/xml: update the Zhaoxin KH-40000 cpuid test (details)
1009. x86: update topology enumeration loop break-out conditions (details)
1010. tests/x86: add new KH-40000 cpuid tarball (details)
1011. linux: try standard cgroup/cpuset mount point before reading /proc/mounts (details)
1012. utils/info: small improvements to manpage and usage (details)
1013. utils/info: abstract-out functions to show levels, topology info and support (details)
1014. utils/info: "levels", "topology" and "support" may now be passed as target objects (details)
1015. utils/info/tests: switch to new keywords "levels", "support" and "topology" (details)
1016. utils/info: --topology, --support and --objects deprecated in 3.0 (details)
1017. utils/info: abstract out the printing of attribute lines (details)
1018. utils/info: add --get-attr to output a single attribute without prefix etc (details)
1019. core: export remove_infos() (details)
1020. memattrs: remove MemoryTier info when not valid anymore (details)
1021. memattrs: add MemoryTiersNr=2 info attr to topology (details)
1022. doxy: typo (details)
1023. contrib/ci.inria.fr: only keep last 30 builds and last 30 days (details)
1024. contrib/ci.inria.fr: actually keep last 25 builds for max 50 days (details)

doxy: more links from/to the topo attributes section

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

completions: add info/calc/bind/distrib --restrict-flags

Forgotten in a7a9603a8e3fa4b064e73e539fb109c0d32428fb

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/dist/publish_doc: display home page links

We often have to update home page links to examples/upgradeAPI pages,
have the doc install script list them.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: add support for generic-initiator ACPI HMAT information

Starting with Linux 5.10, the kernel may expose Generic Initiators
(could be a FPGA/GPU) that have faster access to some memory targets
than usual CPU initiators.

nodes' access0 sysfs directory now contains the fastest of GI and
CPU initiators. access1 contains the fastest of CPU initiators only
(just like access0 in previous kernels).

Hence we now try access1 to avoid GI if any, or fallback to
access0 otherwise (earlier kernels).

Older hwloc releases without knowledge of GI/access1 will
work fine as long as the hardware doesn't expose a GI
(unlikely in the near future).
If the hardware exposes some GIs, we may fail to find
the CPU-locality of some memory targets.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

Move PGI compiler check before GNU

 * PGI compiler has a GNU compatability mode so `__GNUC__` is defined by
   `pgcc`. So first check for `__PGI` before checking for GNU.

Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>

linux: fix comments about generic initiators

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: add support for generic-initiator ACPI HMAT performance information

Commit ac6a1381e99510792393b7ba927f4caa69ea663a was about locality,
this one is for latency/bandwidth memattrs.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/linux: add a sysfs test with complex initiators/targets

There's one node per pair of core and one per 4-core package.

All nodes are either cpu-only (0-3) or memory-only (5-10)
Hence we have to hide CPU-only nodes and find the locality of memory nodes
using sysfs target/initiator files.

There's also a generic initiator (node4) that has faster access
to memory node5/6/7 than its local CPU.
It doesn't appear in our topology but forces us to use access1 instead of access0.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/linux: remove fakememinitiators-1npc+1npc

Its name was wrong (1npc+1npc instead of 1npc+1npp),
and the new fakememinitiators-1np2c+1npp+gi added earlier
covers more features and corner cases.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/android: fix the VERSION

That version is only used internally, e.g. it appears in generated XMLs.
The apk version comes from the android-specific files.

Signed-off-by: Valentin Hoyet <valentin.hoyet@inria.fr>

contrib/android: also fix version individual major/minor/release numbers

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/android: also update the apk version

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/android: remove HWLOC_VERSION from private config.h

The one is the public one is enough,
having to manually update a single one is better.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/git-hooks: add a script for checking version numbers in pre-commit hooks

Android and Windows have their own custom config.h files with
hardwired version numbers. We always forget to update them
(that's why contrib/windows/test-windows-version.sh fails in CI after releases,
 but checking before commit/push would be better).
Android also requires to update the apk version in the gradle config.

We add a check-versions.sh script (which verifies both windows and
android version numbers) and may be installed in a git pre-commit hook.

Signed-off-by: Valentin Hoyet <valentin.hoyet@inria.fr>
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/git-hooks: add a pre-commit hook template

It calls check-versions.sh and update-my-copyright.pl
It must be installed in .git/hooks/pre-commit
(and/or in ~/.git_template/hooks/pre-commit for future clones).
Not sure we can enforce this installation, but we don't want anyway
since people may want to add more checks there.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
signed-off-by: Valentin Hoyet <valentin.hoyet@inria.fr>

bitmap: fix hwloc_bitmap_nr_ulongs()

Thanks to Norbert Eicker for the report.

Closes #429

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/bitmap: check bitmaps with single 0 bit and single full ulong

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/ci.inria.fr: display the windows MSVC build environment

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/ci.inria.fr: run hwloc-info --support on cygwin too

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

lstopo/cairo: only set the line thickness once

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/android: update to apk 1.5.1

- Add link to the discovery debug/log file
  (instead of having to dig into android storage manually)
- Add about page with links to website, github, CI
- Random small improvements

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/android: cleanup RGB color parameter names

No functional change, just RGB instead of RBG in some lines.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

git-hooks/check-version: update to new android gradle version format

Signed-off-by: Valentin Hoyet <valentin.hoyet@inria.fr>

lstopo-ascii: Use last colors instead of first colors

People would be used to defining their own colors, typically starting
from color 16. When we know max_value from ncurses, lstopo should rather
define colors from last colors, to avoid clashing with user-defined
colors.

Fixes #388

memattrs: try to clarify that most initiators should be cpusets for now

Objects as initiators work but it's not used internally yet.

Refs #433

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

memattrs: some doxygen fixes

Hide initializers,
Move enum declarations out of structs so that references don't need the struct prefix.
Misc.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: factorize some fd/path reading functions back

There's no strong need to have read_fd_as_cpumask/list()
separated from read_path_as_cpumask/list(), merge them back.

Also it avoids using open() in some cases (no fsroot) and
hwloc_open() in others (fsroot).

Also it will simplify support for readfile() (if it ever gets merged)
since we now open/read/close in the same function, but there's still
some work to do for reading cpumask/cpulist with readfile().

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: rework read_path_by_length() return value

Return the length read instead of success/failure.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: use helpers for reading tid location

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: use helpers for reading nvidia gpu numa nodes

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: use helpers for reading os device numa node

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: use helpers for reading the pci config space

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/diff: fix the test when the first NUMA node is CPU less

NUMA differences may actually appear before PU differences if
the first NUMA node is CPU less.
Handle both orders in the assertions.

Thanks to James D. Trotter for reporting the issue.
Closes #374.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

core: abstract-out hwloc__tma_dup_infos()

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

API: add hwloc/cpukinds.h for exposing difference between cores

Backends may register cpusets with optional info attributes
describing contained PUs.
The core will agregate these info into individual "kinds" of
PUs with same attributes.
Each PU may be part of a single kind.

Backends may also force an efficiency value that the core will
use to rank kinds.

We'll later add heuristics for ranking kinds based on info attributes.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

xml/import: factorize out the actual parsing on non-object-specific info attr

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

xml: add cpukinds import/export

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/xml: add a case with cpukinds

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

lstopo: add --cpukinds option

They are also shown in verbose mode.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

x86: use the hybrid leaf to build a cpukinds based on the core type

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: read cpufreq/cpuinfo_max_freq and base_frequency to create cpukinds

cpuinfo_max_freq is available in all cpufreq drivers,
but it's not totally clear it matches efficiency.

base_frequency is likely more reliable,
but only available in the intel_pstate cpufreq driver.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

windows: use RtlGetVersion() for reliable version checks without manifest

GetVersion() works fine inside hwloc now that we build with a manifest.
However, external users will need a manifest too when building their
program so that their hwloc enables Windows10+ code properly (even if
they use a libhwloc that was built with a manifest). Not easy.

Use the RtlGetversion() symbol to get the real Windows version.
This is sort of ugly but many people have used this for a while,
hence Microsoft will likely have a hard time breaking it.
If the symbol is ever missing, we'll fallback to manifest+GetVersion.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

windows: add cpukinds using the native efficiency class in Windows 10

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

cpukinds: add some ranking/efficiency heuristics

We first look at the forced efficiency if any (provided by Windows 10 only for now).
Otherwise, we use the coretype (IntelAtom or IntelCore) and frequencies
to define efficiencies (base frequency if available, max frequency otherwise).
On ARM, we only use frequencies since there is no information about
core type.

It's not clear how things will evolve in the future, so we have several
other heuristics that may be enbaled through the HWLOC_CPUKINDS_RANKING
env var (coretype+frequency/coretype/frequency/frequency_max/frequency_base/
forced_efficiency/no_forced_efficiency/none).

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

cpukinds: deduplicate identical info names/values

May happen when both Linux and x86 report CoreType

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

doxy: add cpukinds info in topology attributes

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/linux: add heterogenous frequencies to fakeheteronuma for cpukinds

And rename it to fakeheterocpunuma

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/x86: add a test with fake cpukind information

Modified from my laptop by adding 0x1a CPUID leaf exposing Intel Atom/Core types,
and by setting the hybrid feature (bit 15 of cpuid7.edx).

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

utils/hwloc-calc: add --cpukind

For filtering PUs by cpu kind

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

utils/tests: check cpukinds in test-hwloc-calc

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

utils/hwloc-info: display which CPU kind(s) CPU-side objects are included in

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

utils/tests: check hwloc-info cpukinds output

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

API: add hwloc_cpukinds_register()

For adding custom CPU kind information.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests: add a cpukinds API test

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

utils/hwloc-annotate: add cpukind command

For adding a new cpukind

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

utils/tests: check cpukinds in hwloc-annotate

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

API: add a cpukind_efficiency bit in the discovery support array

Only set on Windows for now.
Forced efficiency are only available on Windows, or when given by the user.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

lstopo/core: add cpukind attributes to object userdata

Not used yet, except for disabling the factorizing across different kinds.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

lstopo/draw: change the format of PU boxes for CPU kinds

When there are multiple kinds, use dot/dash lines with
higher thickness for PUs not of the first kind.
Bold font is also used for kinds with odd indexes.

Kind#0 gets normal boxes
Kind#1 gets boxes with small dashes and bold font
Kind#2 gets larger dashes, thicker lines, and non-bold font
etc

The thickness isn't the same in all backends,
depends on the actual rendering.
Windows doesn't support dash/dot if thickness>1,
hence we don't change the thickness at all there.

All this may be disabled with --no-cpukinds,
or toggled with 'k' in the interactive outputs.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

Thanks to Clement Foyer for the tikz part.

Signed-off-by: Clement Foyer <cfoyer@hpe.com>

linux/cpukinds: maxfreq isn't always maxfreq anymore

Can be either max ou base freq.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux/cpukinds: sort by frequency before registering

Lower frequency likely means lower efficiency.
Hence register frequency-based cpukinds in that order for
consistency with cases where we have enough info for ranking.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

API: bump HWLOC_API_VERSION to 0x00020400

hwloc/cpukinds.h was added.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

ci.inria.fr/browse_jenkins_logs.sh: update examples

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/android: fix the apk manifest name

Likely got wrong during a merge conflict.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/android: save the app icon original files

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/android: fix sharing of graphic output

Share the entire lstopo output, not the screen.

Signed-off-by: Valentin Hoyet <Valentin.Hoyet@inria.fr>

linux: use Linux 5.11 sysfs cpu types to detect Intel Atom/Core types

It's not clear it will be in 5.11 or not,
and whether the sysfs files/directories will be different.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/linux: add Linux sysfs cpu types to fakeheterocpunuma

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

doxy: fix LinuxCPUTyle cpukind attribute description

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

VERSION: master is now 2.5

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

cpukinds: fix the checking for duplicate values when ranking by info

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

distances: uniformize and document NUMALatency matrices

Use the same name for Linux/Solaris/FreeBSD,
and cite it as the main example in the doc.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

autogen: fix MacOS X Big Sur libtool.m4 bug by patching configure

This libtool patch isn't applied upstream yet but it is
already used by several projects to fix build failures
on the latest MacOS X (Big Sur).

Upstream patch at https://lists.gnu.org/archive/html/libtool-patches/2020-06/msg00001.html
applies to libtool.m4 but we have to apply it to configure
after autoconf used the (buggy) libtool.m4 instead.

See also open-mpi/ompi#8218.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

utils/misc: remove a useless and harmless loop duplicate

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

utils/misc: improve the printing of distance matrices

Scale the column width to the max of what we have to show in the matrix,
don't limit values to 5 digits, etc.

Also print OSDevices by name instead of indexes.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

core: fortify bridge downstream attribute use

All our bridges are currently downstream to PCI,
but it's not impossible it changes in the future,
make the code a bit more safe.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

git-hooks/check-versions: better error messages when Windows VERSION is outdated

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

NEWS: bring official 2.4.0 bullets in master

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

Fixed a memory leak when destroying hwloc_linux_cpufreqs objects.

Signed-off-by: Perceval Anichini <perceval.anichini@trilogic.fr>

distances: fix the import of large heterogeneous distance matrices

Properly import the list of heterogeneous indexes when it is split
on multiple lines.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/xml/fakeheterodistances: enlarge matrices to test hetero-indexes on multiple lines

Test for commit e7bf9720fade2c1924cd55a90ece0590a46c0246

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

plugins.h: clarify that things may change when the component ABI is bumped

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

ci.inria.fr/sonar: update languages for recent sonarscanner releases

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

ci.inria.fr/sonar: output the sonar-scanner version

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

configure: add --with-cuda=<dir>

Specifies where CUDA is installed to avoid having to pass
C_INCLUDE_PATH and LIBRARY_PATH at build time.
LD_LIBRARY_PATH is still required at runtime.

The CUDA directory is also used for looking for OpenCL and NVML
subcomponents.

Thanks to Andrea Bocci for the suggestion.

Refs #392.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

hwloc/opencl.h: fix AMD OpenCL device locality when PCI bus/device >= 128

cl_device_topology_amd stores the PCI bus/device/function as cl_char (signed).
bus/device can be >= 128 and those don't cast correctly to unsigned int.

Applications that call the inline function hwloc_opencl_get_device_pci_busid()
directly must be recompiled to get this fix.

Thanks to Edgar Leon for reporting the issue (some AMD OpenCL devices were
not inserted in the PCI parent while RSMI devices were fine).

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

ci.inria.fr/sonarscanner: update exclusions

Fix tests/ports exclusion
Add XMLs (we already verify them with xmllint and lstopo)
Add netloc_draw (we don't want to test all this HTML/JS code)

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

Fix GCC checks to use the vendor name

 * The PGI compiler is identified as a GCC compiler (so `$GCC == yes`)
   This can cause problems if the user accidentally adds `--enable-picky`
   or if the `$GCC` variable is used to check if it is safe to add GCC
   specific compiler flags.
 * Instead of using `$GCC` we can use the vendor string returned by
   `_HWLOC_C_COMPILER_VENDOR`.

Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>

contrib/android: remove unused duplicate android manifest

Wasn't correctly fixed in a55490d4f65bfc97033b49a5ae237943e28d5cc7

Thanks to Eric Schnetter for reporting the issue,
it caused a filename conflict on case-insensitive filesystem (MacOS).

Closes #441

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

x86: add a summary of PU/APIC/Core/Pkg/Node info in verbose debug

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

x86: fix full discovery enabling when FreeBSD/NetBSD backends only created PUs

Backends such as FreeBSD and NetBSD only create PUs but their
priority is higher than x86. Hence x86 has to detect whether only
PUs were created before adding everything else.
This requires to reconnect levels earlier.

Thanks to Thibault Payet and Yuri Victorovich for the report.

Closes #442

Downstream FreeBSD bug report https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=252643
HPX issue about requiring Core objects https://github.com/STEllAR-GROUP/hpx/issues/5115

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

x86: update general comments about this backend

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

doc: Add missing dependency

Signed-off-by: Samuel Thibault <samuel.thibault@ens-lyon.org>

configure: pass -Werror when checking compiler options

clang warns about unknown options (while gcc errors-out)
so pass -Werror when checking for options at configure time.

Thanks to Jeff Squyres for the report.
Refs #447

By the way, rename _HWLOC_CHECK_GCC_OPTION() to _HWLOC_CHECK_CC_OPTION()
since this isn't GCC specific.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

autogen: use bash explicitly

echo -n doesn't work on some Mac OSX

Refs #447

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

configure: detect when a patched libtool was used

Some libtool are now available with a Big Sur patch.
Detect both versions of the patch before trying to patch.

Thanks to Jeff Squyres for the report.
Closes #447

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

core: mark the topology as modified when we replace an object

When inserting a custom group after load(), it may get merged
and replace an existing group. This requires a reconnect.
So mark the topology as modified so that levels are rebuilt, etc.

Refs #446

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

core: fix inserting of custom groups replacing an existing ones

When inserting a group identical to an existing one,
the new one may "replace" the old one (depends on their kinds).
If so, we need to rebuild the level, fill its cpusets,
recompute its depth and total_memory, etc.

This happened when lstopo --top inserted a group (kind 0 for now)
identical to a NUMA group (kind higher).

Thanks to Scott Hicks for reporting the issue and providing
lots of debugging information.

Closes #446

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/groups: check the merging of identical groups with higher/smaller kinds

Reproducer for #446

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

API: clarify the documentation of alloc/insert_group()

Detail the behavior of merging with respect to dont_merge and kinds.
Allow to use high kinds for discarding whenever possible, etc.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

lstopo: hide --top groups whenever possible

Set their kinds to max so that they don't replace existing
memory/... kinds if identical.

Refs #446.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

core: add assertions about local/total memory in debug

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

utils/hwloc-ps.1: fix a ref to --disallowed

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/shmem: cleanup the enabling of specific distance checks

Add a custom info attr instead on relaying on the existing
SyntheticDescription (which will exist in some upcoming XML
topology tests).

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/shmem: also test topologies with memattrs/cpukinds/io from XML

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

Fix check for IBM xl compilers for v13.1 and later.

By default newer xlc compilers only define __ibmxl__ now.

https://www.ibm.com/support/knowledgecenter/en/SSXVZZ_13.1.6/com.ibm.xlcpp1316.lelinux.doc/compiler_ref/xlmacros.html
Signed-off-by: Austen Lauria <awlauria@us.ibm.com>

Fix configury where most compilers will get mislabeled as 'gnu'.

Most compilers define the __GNU__ macro, so put it at the bottom
as a catch-all.

Tested with gcc, xlc, pgi and clang.

Signed-off-by: Austen Lauria <awlauria@us.ibm.com>

utils/gather-cpuid: dump the 0x19 leaf

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

configure: don't define cuda common vars unless needed (opencl/nvml/cuda)

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

cpukinds: fix the checking for duplicate ranking values

Don't compare forced_efficiency, compare ranking values instead.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/x86+linux: allow .env file to change the x86,linux order

Testing the default order (linux before x86) will be useful too.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/x86+linux: add a test combining x86 CoreType and linux cpufreq

Based on Intel Lakefield.

Regression tests for commit 9cd818b7715dc48d0610e05cdafc30e7625af229

x86 must be called after linux because linux cannot annotate cpufreqs yet.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests/x86+linux: don't forget the new options file in make dist

Forgotten in 1a97da7a02b6e851229082720ec4bd7b684962c2

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: allow annotating cpukinds when PUs were discovered by another backend (x86)

Will be useful when x86 must run first because the CPUs
isn't correctly exposed by Linux.

This behavior is now tested in 64amd64-4p2n4ca2co which now
gets cpukinds frequency info.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: fix support for CPU0 being hotplugged

Linux kernel with CONFIG_BOOTPARAM_HOTPLUG_CPU0 as compile time
allow offlining CPU0. The Linux early discovery used CPU0 to detect
whether we should use /sys/bus/cpu or /sys/devices/system/cpu,
and whether new core/package_cpus are available.
If this fails, read the first online CPU number at look at it.
The overhead for the normal case (CPU0 online) is 0.

Thanks to Garrett Clay for the report.
Fixes #450

For hwloc 2.5, we should moderniez this code:
/sys/devices/system/cpu is unlikely going away because many
files were added in there since /sys/bus/cpu was added.
Kernels without sysfs cpu topology files are also too old now.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

hwloc-ps: add --short-name

Signed-off-by: Valentin Hoyet <valentin.hoyet@inria.fr>

hwloc-ps: don't apply --pid-cmd to processes that will be filtered out

Split the calling of the user-provided command out of the main
common-ps functions and let the caller calls it after filtering
interesting processes.

Otherwise hwloc-ps --pid-cmd takes several seconds on all platforms
since we have to query hundreds of (uninteresting because not bound)
processes.

Thanks to Valentin Hoyet for the report.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

cuda/nvml/opencl/pci/rsmi: prefix error messages with "hwloc"

Refs #453

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

utils/common-ps: fix popen/pclose for Windows

Broken in cbb6b960f9c1d5c2cc205e3a3df78454bd551d13

Not strictly needed since hwloc-ps isn't built on Windows
and lstopo --ps does nothing there, but the build is easy to fix.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

configure: fix cuda common variable definition

Fix for 29e3ec3afc1624ec012519c519dbb5ab277d6054

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

NEWS: bring 2.4.1 bullets in master

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

utils/common-ps: refix popen/pclose for Windows/Cygwin

Completes commit 5fd60f712f4fe52b07ee54e738ee76288a5c7111,
which tried to fix cbb6b960f9c1d5c2cc205e3a3df78454bd551d13

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

configure: move AC_ARG_VAR(MS_LIB) to hwloc-internal config

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

configure: typo in comments

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

doxy: add Cygwin to the FAQ about Windows builds

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

configure: cleanup cuda/cudart variables

libhwloc uses cudart only. cuda is for tests/hwloc/cuda.
Rename configure/Makefile variables properly.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

configure: use LIBS variables for ibverbs/rsmi/nvml tests build

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

configure: use CPPFLAGS/CFLAGS/LIBS/LDFLAGS consistently

Some of these variables aren't defined for some backends,
depending on whether we detect using pkg-config or not.
Don't propagate these differences everywhere, always use
CPPFLAGS/CFLAGS/LIBS/LDFLAGS even if they are empty for now.

Applies to global lib build variables, plugin variables,
and tests variables.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

configure: add --with-cuda-version or use CUDA_VERSION to select CUDA pkg-config files

Thanks to Stephen Herbein for the suggestion.

Refs #392

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

doxy: add a FAQ entry about CUDA configuration

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

cuda: don't warn if no device were found

cuda returns an error when there are no devices.

Unfortunately, nvmlInit() returns "unknown error" (999)
in the same case. I am not sure we can filter-out this
error in all 999 cases.

Thanks to Aurelien Bouteiller for the report.

Refs #453

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

API: add hwloc/windows.h for querying windows processor groups

Windows processor groups are not a physical hierarchical level
(although it may be exposed by hwloc if it's different from
 NUMA nodes, packages, etc) but it has an impact of application
because binding must remain inside a single group.
Expose those groups explicitly so that applications know what
their binding masks should be restricted to.

Thanks to Ivan Kochin for the suggestion.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

tests: add a test for the windows processor group API

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

lstopo: add --windows-processor-groups

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

doxy: add links from interoperability items to dedicated API sections

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

windows: fix a leak in the initialization of processor group cpusets

CID#1473456

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

windows: properly destroy the processor group cpuset array

CID#1473455

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

lstopo: fix --windows-processor-groups name confusion

--processor-groups was only used during development,
real/final name is --windows-processor-groups

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/ci.inria.fr: add --windows-processor-groups to windows lstopo outputs

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

core: don't display "less critical" error messages by default

Failure to initialize CUDA isn't critical enough to pollute
external users' stacks with hwloc warnings.
Set HWLOC_HIDE_ERRORS to 1 by default to that those aren't
displayed by default anymore.
Set it back to 0 in lstopo to get them there.

Invalid hardware topology, invalid XML input, missing 32bits
PCI domains, etc. are much more important, display them unless
HWLOC_HIDE_ERRORS is set to 2.

Thanks to Aurelien Bouteiller in #453.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

components: cleanup error messages

Make them hidable through HWLOC_HIDE_ERRORS
and prefix all messages (errors or verbose) with "hwloc: ".

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

AIX: cleanup error messages

Make them hidable through HWLOC_HIDE_ERRORS
and prefix them with "hwloc/aix: ".

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

solaris: cleanup error messages

Make them hidable through HWLOC_HIDE_ERRORS
and prefix them with "hwloc/solaris: ".

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

cpukinds: prefix the error message with "hwloc: "

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

pci: cleanup some error messages

Make them hidable through HWLOC_HIDE_ERRORS
and prefix them with "hwloc/pci: ".

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux/knl: cleanup error messages

Make them hidable through HWLOC_HIDE_ERRORS
and prefix them with "hwloc/linux: ".

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: cleanup more error messages

Make them hidable through HWLOC_HIDE_ERRORS
and prefix them with "hwloc/linux: ".

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux+xml: cleanup some error messages about 32bits PCI domain

Make them hidable through HWLOC_HIDE_ERRORS
and prefix them with "hwloc/foo: ".

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

xml: add a missing verbose condition before a printf

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

xml: cleanup some error messages

Make them hidable through HWLOC_HIDE_ERRORS
and prefix them with "hwloc/xml: ".

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

core: cleanup error messages

Make them hidable through HWLOC_HIDE_ERRORS
and prefix them with "hwloc: ".

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

NEWS: reorder items

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

rsmi: also limit rsmi shutdown calls on error

Forgotten in 619cbcd17cb5a1b6fef2b1f8b736100cd597cf2e

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

hwloc.ps: add --single-ancestor option

Signed-off-by: Valentin Hoyet <valentin.hoyet@inria.fr>
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

contrib/completion: add lstopo --windows-processor-groups

Forgotten in 588a4346d2cd9e3bd9a2a9ef02266323fbfc666f

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>