v6.23.0-dev

Better detection of aligned malloc functions

.github/ISSUE_TEMPLATE/bug_report.md

@@ -17,6 +17,9 @@ Steps to reproduce the behavior.
 A clear and concise description of what you expected to happen.
 
 **Required data**
+ - XMRig version
+    - Either the exact link to a release you downloaded from https://github.com/xmrig/xmrig/releases
+    - Or the exact command lines that you used to build XMRig
  - Miner log as text or screenshot
  - Config file or command line (without wallets)
  - OS: [e.g. Windows]

.gitignore

@@ -1,4 +1,6 @@
 /build
+scripts/build
+scripts/deps
 /CMakeLists.txt.user
 /.idea
 /src/backend/opencl/cl/cn/cryptonight_gen.cl

CHANGELOG.md

@@ -1,3 +1,286 @@
+# v6.22.3
+- [#3605](https://github.com/xmrig/xmrig/pull/3605) CUDA backend: added missing RandomX dataset update.
+- [#3646](https://github.com/xmrig/xmrig/pull/3646) Optimized auto-config for AMD CPUs with less than 2 MB L3 cache per thread.
+- [#3652](https://github.com/xmrig/xmrig/pull/3652) Fixed possible crash when submitting RandomX benchmark.
+- [#3662](https://github.com/xmrig/xmrig/pull/3662) Fixed OpenCL kernel compilation error on some platforms.
+
+# v6.22.2
+- [#3569](https://github.com/xmrig/xmrig/pull/3569) Fixed corrupted API output in some rare conditions.
+- [#3571](https://github.com/xmrig/xmrig/pull/3571) Fixed number of threads on the new Intel Core Ultra CPUs.
+
+# v6.22.1
+- [#3531](https://github.com/xmrig/xmrig/pull/3531) Always reset nonce on RandomX dataset change.
+- [#3534](https://github.com/xmrig/xmrig/pull/3534) Fixed threads auto-config on Zen5.
+- [#3535](https://github.com/xmrig/xmrig/pull/3535) RandomX: tweaks for Zen5.
+- [#3539](https://github.com/xmrig/xmrig/pull/3539) Added Zen5 to `randomx_boost.sh`.
+- [#3540](https://github.com/xmrig/xmrig/pull/3540) Detect AMD engineering samples in `randomx_boost.sh`.
+
+# v6.22.0
+- [#2411](https://github.com/xmrig/xmrig/pull/2411) Added support for [Yada](https://yadacoin.io/) (`rx/yada` algorithm).
+- [#3492](https://github.com/xmrig/xmrig/pull/3492) Fixed `--background` option on Unix systems.
+- [#3518](https://github.com/xmrig/xmrig/pull/3518) Possible fix for corrupted API output in rare cases.
+- [#3522](https://github.com/xmrig/xmrig/pull/3522) Removed `rx/keva` algorithm.
+- [#3525](https://github.com/xmrig/xmrig/pull/3525) Added Zen5 detection.
+- [#3528](https://github.com/xmrig/xmrig/pull/3528) Added `rx/yada` OpenCL support.
+
+# v6.21.3
+- [#3462](https://github.com/xmrig/xmrig/pull/3462) RandomX: correct memcpy size for JIT initialization.
+
+# v6.21.2
+- The dependencies of all prebuilt releases have been updated. Support for old Ubuntu releases has been dropped.
+- [#2800](https://github.com/xmrig/xmrig/issues/2800) Fixed donation with GhostRider algorithm for builds without KawPow algorithm.
+- [#3436](https://github.com/xmrig/xmrig/pull/3436) Fixed, the file log writer was not thread-safe.
+- [#3450](https://github.com/xmrig/xmrig/pull/3450) Fixed RandomX crash when compiled with fortify_source.
+
+# v6.21.1
+- [#3391](https://github.com/xmrig/xmrig/pull/3391) Added support for townforge (monero fork using randomx).
+- [#3399](https://github.com/xmrig/xmrig/pull/3399) Fixed Zephyr mining (OpenCL).
+- [#3420](https://github.com/xmrig/xmrig/pull/3420) Fixed segfault in HTTP API rebind.
+
+# v6.21.0
+- [#3302](https://github.com/xmrig/xmrig/pull/3302) [#3312](https://github.com/xmrig/xmrig/pull/3312) Enabled keepalive for Windows (>= Vista).
+- [#3320](https://github.com/xmrig/xmrig/pull/3320) Added "built for OS/architecture/bits" to "ABOUT".
+- [#3339](https://github.com/xmrig/xmrig/pull/3339) Added SNI option for TLS connections.
+- [#3342](https://github.com/xmrig/xmrig/pull/3342) Update `cn_main_loop.asm`.
+- [#3346](https://github.com/xmrig/xmrig/pull/3346) ARM64 JIT: don't use `x18` register.
+- [#3348](https://github.com/xmrig/xmrig/pull/3348) Update to latest `sse2neon.h`.
+- [#3356](https://github.com/xmrig/xmrig/pull/3356) Updated pricing record size for **Zephyr** solo mining.
+- [#3358](https://github.com/xmrig/xmrig/pull/3358) **Zephyr** solo mining: handle multiple outputs.
+
+# v6.20.0
+- Added new ARM CPU names.
+- [#2394](https://github.com/xmrig/xmrig/pull/2394) Added new CMake options `ARM_V8` and `ARM_V7`.
+- [#2830](https://github.com/xmrig/xmrig/pull/2830) Added API rebind polling.
+- [#2927](https://github.com/xmrig/xmrig/pull/2927) Fixed compatibility with hwloc 1.11.x.
+- [#3060](https://github.com/xmrig/xmrig/pull/3060) Added x86 to `README.md`.
+- [#3236](https://github.com/xmrig/xmrig/pull/3236) Fixed: receive CUDA loader error on Linux too.
+- [#3290](https://github.com/xmrig/xmrig/pull/3290) Added [Zephyr](https://www.zephyrprotocol.com/) coin support for solo mining.
+
+# v6.19.3
+- [#3245](https://github.com/xmrig/xmrig/issues/3245) Improved algorithm negotiation for donation rounds by sending extra information about current mining job.
+- [#3254](https://github.com/xmrig/xmrig/pull/3254) Tweaked auto-tuning for Intel CPUs.
+- [#3271](https://github.com/xmrig/xmrig/pull/3271) RandomX: optimized program generation.
+- [#3273](https://github.com/xmrig/xmrig/pull/3273) RandomX: fixed undefined behavior.
+- [#3275](https://github.com/xmrig/xmrig/pull/3275) RandomX: fixed `jccErratum` list.
+- [#3280](https://github.com/xmrig/xmrig/pull/3280) Updated example scripts.
+
+# v6.19.2
+- [#3230](https://github.com/xmrig/xmrig/pull/3230) Fixed parsing of `TX_EXTRA_MERGE_MINING_TAG`.
+- [#3232](https://github.com/xmrig/xmrig/pull/3232) Added new `X-Hash-Difficulty` HTTP header.
+- [#3240](https://github.com/xmrig/xmrig/pull/3240) Improved .cmd files when run by shortcuts on another drive.
+- [#3241](https://github.com/xmrig/xmrig/pull/3241) Added view tag calculation (fixes Wownero solo mining issue).
+
+# v6.19.1
+- Resolved deprecated methods warnings with OpenSSL 3.0.
+- [#3213](https://github.com/xmrig/xmrig/pull/3213) Fixed build with 32-bit clang 15.
+- [#3218](https://github.com/xmrig/xmrig/pull/3218) Fixed: `--randomx-wrmsr=-1` worked only on Intel.
+- [#3228](https://github.com/xmrig/xmrig/pull/3228) Fixed build with gcc 13.
+
+# v6.19.0
+- [#3144](https://github.com/xmrig/xmrig/pull/3144) Update to latest `sse2neon.h`.
+- [#3161](https://github.com/xmrig/xmrig/pull/3161) MSVC build: enabled parallel compilation.
+- [#3163](https://github.com/xmrig/xmrig/pull/3163) Improved Zen 3 MSR mod.
+- [#3176](https://github.com/xmrig/xmrig/pull/3176) Update cmake required version to 3.1.
+- [#3182](https://github.com/xmrig/xmrig/pull/3182) DragonflyBSD compilation fixes.
+- [#3196](https://github.com/xmrig/xmrig/pull/3196) Show IP address for failed connections.
+- [#3185](https://github.com/xmrig/xmrig/issues/3185) Fixed macOS DMI reader.
+- [#3198](https://github.com/xmrig/xmrig/pull/3198) Fixed broken RandomX light mode mining.
+- [#3202](https://github.com/xmrig/xmrig/pull/3202) Solo mining: added job timeout (default is 15 seconds).
+
+# v6.18.1
+- [#3129](https://github.com/xmrig/xmrig/pull/3129) Fix: protectRX flushed CPU cache only on MacOS/iOS.
+- [#3126](https://github.com/xmrig/xmrig/pull/3126) Don't reset when pool sends the same job blob.
+- [#3120](https://github.com/xmrig/xmrig/pull/3120) RandomX: optimized `CFROUND` elimination.
+- [#3109](https://github.com/xmrig/xmrig/pull/3109) RandomX: added Blake2 AVX2 version.
+- [#3082](https://github.com/xmrig/xmrig/pull/3082) Fixed GCC 12 warnings.
+- [#3075](https://github.com/xmrig/xmrig/pull/3075) Recognize `armv7ve` as valid ARMv7 target.
+- [#3132](https://github.com/xmrig/xmrig/pull/3132) RandomX: added MSR mod for Zen 4.
+- [#3134](https://github.com/xmrig/xmrig/pull/3134) Added Zen4 to `randomx_boost.sh`.
+
+# v6.18.0
+- [#3067](https://github.com/xmrig/xmrig/pull/3067) Monero v15 network upgrade support and more house keeping.
+  - Removed deprecated AstroBWTv1 and v2.
+  - Fixed debug GhostRider build.
+  - Monero v15 network upgrade support.
+  - Fixed ZMQ debug log.
+  - Improved daemon ZMQ mining stability.
+- [#3054](https://github.com/xmrig/xmrig/pull/3054) Fixes for 32-bit ARM.
+- [#3042](https://github.com/xmrig/xmrig/pull/3042) Fixed being unable to resume from `pause-on-battery`.
+- [#3031](https://github.com/xmrig/xmrig/pull/3031) Fixed `--cpu-priority` not working sometimes.
+- [#3020](https://github.com/xmrig/xmrig/pull/3020) Removed old AstroBWT algorithm.
+
+# v6.17.0
+- [#2954](https://github.com/xmrig/xmrig/pull/2954) **Dero HE fork support (`astrobwt/v2` algorithm).**
+  - [#2961](https://github.com/xmrig/xmrig/pull/2961) Dero HE (`astrobwt/v2`) CUDA config generator.
+  - [#2969](https://github.com/xmrig/xmrig/pull/2969) Dero HE (`astrobwt/v2`) OpenCL support.
+- Fixed displayed DMI memory information for empty slots.
+- [#2932](https://github.com/xmrig/xmrig/pull/2932) Fixed GhostRider with hwloc disabled.
+
+# v6.16.4
+- [#2904](https://github.com/xmrig/xmrig/pull/2904) Fixed unaligned memory accesses.
+- [#2908](https://github.com/xmrig/xmrig/pull/2908) Added MSVC/2022 to `version.h`.
+- [#2910](https://github.com/xmrig/xmrig/issues/2910) Fixed donation for GhostRider/RTM.
+
+# v6.16.3
+- [#2778](https://github.com/xmrig/xmrig/pull/2778) Fixed `READY threads X/X` display after algorithm switching.
+- [#2782](https://github.com/xmrig/xmrig/pull/2782) Updated GhostRider documentation.
+- [#2815](https://github.com/xmrig/xmrig/pull/2815) Fixed `cn-heavy` in 32-bit builds.
+- [#2827](https://github.com/xmrig/xmrig/pull/2827) GhostRider: set correct priority for helper threads.
+- [#2837](https://github.com/xmrig/xmrig/pull/2837) RandomX: don't restart mining threads when the seed changes.
+- [#2848](https://github.com/xmrig/xmrig/pull/2848) GhostRider: added support for `client.reconnect` method.
+- [#2856](https://github.com/xmrig/xmrig/pull/2856) Fix for short responses from some Raptoreum pools.
+- [#2873](https://github.com/xmrig/xmrig/pull/2873) Fixed GhostRider benchmark on single-core systems.
+- [#2882](https://github.com/xmrig/xmrig/pull/2882) Fixed ARMv7 compilation.
+- [#2893](https://github.com/xmrig/xmrig/pull/2893) KawPow OpenCL: use separate UV loop for building programs.
+
+# v6.16.2
+- [#2751](https://github.com/xmrig/xmrig/pull/2751) Fixed crash on CPUs supporting VAES and running GCC-compiled xmrig.
+- [#2761](https://github.com/xmrig/xmrig/pull/2761) Fixed broken auto-tuning in GCC Windows build.
+- [#2771](https://github.com/xmrig/xmrig/issues/2771) Fixed environment variables support for GhostRider and KawPow. 
+- [#2769](https://github.com/xmrig/xmrig/pull/2769) Performance fixes:
+  - Fixed several performance bottlenecks introduced in v6.16.1.
+  - Fixed overall GCC-compiled build performance, it's the same speed as MSVC build now.
+  - **Linux builds are up to 10% faster now compared to v6.16.0 GCC build.**
+  - **Windows builds are up to 5% faster now compared to v6.16.0 MSVC build.**
+
+# v6.16.1
+- [#2729](https://github.com/xmrig/xmrig/pull/2729) GhostRider fixes:
+  - Added average hashrate display.
+  - Fixed the number of threads shown at startup.
+  - Fixed `--threads` or `-t` command line option (but `--cpu-max-threads-hint` is recommended to use).
+- [#2738](https://github.com/xmrig/xmrig/pull/2738) GhostRider fixes:
+  - Fixed "difficulty is not a number" error when diff is high on some pools.
+  - Fixed GhostRider compilation when `WITH_KAWPOW=OFF`.
+- [#2740](https://github.com/xmrig/xmrig/pull/2740) Added VAES support for Cryptonight variants **+4% speedup on Zen3**.
+  - VAES instructions are available on Intel Ice Lake/AMD Zen3 and newer CPUs.
+  - +4% speedup on Ryzen 5 5600X.
+
+# v6.16.0
+- [#2712](https://github.com/xmrig/xmrig/pull/2712) **GhostRider algorithm (Raptoreum) support**: read the [RELEASE NOTES](src/crypto/ghostrider/README.md) for quick start guide and performance comparisons.
+- [#2682](https://github.com/xmrig/xmrig/pull/2682) Fixed: use cn-heavy optimization only for Vermeer CPUs.
+- [#2684](https://github.com/xmrig/xmrig/pull/2684) MSR mod: fix for error 183.
+
+# v6.15.3
+- [#2614](https://github.com/xmrig/xmrig/pull/2614) OpenCL fixes for non-AMD platforms.
+- [#2623](https://github.com/xmrig/xmrig/pull/2623) Fixed compiling without kawpow.
+- [#2636](https://github.com/xmrig/xmrig/pull/2636) [#2639](https://github.com/xmrig/xmrig/pull/2639) AstroBWT speedup (up to +35%).
+- [#2646](https://github.com/xmrig/xmrig/pull/2646) Fixed MSVC compilation error.
+
+# v6.15.2
+- [#2606](https://github.com/xmrig/xmrig/pull/2606) Fixed: AstroBWT auto-config ignored `max-threads-hint`.
+- Fixed possible crash on Windows (regression in v6.15.1).
+
+# v6.15.1
+- [#2586](https://github.com/xmrig/xmrig/pull/2586) Fixed Windows 7 compatibility.
+- [#2594](https://github.com/xmrig/xmrig/pull/2594) Added Windows taskbar icon colors.
+
+# v6.15.0
+- [#2548](https://github.com/xmrig/xmrig/pull/2548) Added automatic coin detection for daemon mining.
+- [#2563](https://github.com/xmrig/xmrig/pull/2563) Added new algorithm RandomX Graft (`rx/graft`).
+- [#2565](https://github.com/xmrig/xmrig/pull/2565) AstroBWT: added AVX2 Salsa20 implementation.
+- Added support for new CUDA plugin API (previous API still supported).
+
+# v6.14.1
+- [#2532](https://github.com/xmrig/xmrig/pull/2532) Refactoring: stable (persistent) algorithms IDs.
+- [#2537](https://github.com/xmrig/xmrig/pull/2537) Fixed Termux build.
+
+# v6.14.0
+- [#2484](https://github.com/xmrig/xmrig/pull/2484) Added ZeroMQ support for solo mining.
+- [#2476](https://github.com/xmrig/xmrig/issues/2476) Fixed crash in DMI memory reader.
+- [#2492](https://github.com/xmrig/xmrig/issues/2492) Added missing `--huge-pages-jit` command line option.
+- [#2512](https://github.com/xmrig/xmrig/pull/2512) Added show the number of transactions in pool job.
+
+# v6.13.1
+- [#2468](https://github.com/xmrig/xmrig/pull/2468) Fixed regression in previous version: don't send miner signature during regular mining.
+
+# v6.13.0
+- [#2445](https://github.com/xmrig/xmrig/pull/2445) Added support for solo mining with miner signatures for the upcoming Wownero fork.
+
+# v6.12.2
+- [#2280](https://github.com/xmrig/xmrig/issues/2280) GPU backends are now disabled in benchmark mode.
+- [#2322](https://github.com/xmrig/xmrig/pull/2322) Improved MSR compatibility with recent Linux kernels and updated `randomx_boost.sh`.
+- [#2340](https://github.com/xmrig/xmrig/pull/2340) Fixed AES detection on FreeBSD on ARM.
+- [#2341](https://github.com/xmrig/xmrig/pull/2341) `sse2neon` updated to the latest version.
+- [#2351](https://github.com/xmrig/xmrig/issues/2351) Fixed help output for `--cpu-priority` and `--cpu-affinity` option.
+- [#2375](https://github.com/xmrig/xmrig/pull/2375) Fixed macOS CUDA backend default loader name.
+- [#2378](https://github.com/xmrig/xmrig/pull/2378) Fixed broken light mode mining on x86.
+- [#2379](https://github.com/xmrig/xmrig/pull/2379) Fixed CL code for KawPow where it assumes everything is AMD.
+- [#2386](https://github.com/xmrig/xmrig/pull/2386) RandomX: enabled `IMUL_RCP` optimization for light mode mining.
+- [#2393](https://github.com/xmrig/xmrig/pull/2393) RandomX: added BMI2 version for scratchpad prefetch.
+- [#2395](https://github.com/xmrig/xmrig/pull/2395) RandomX: rewrote dataset read code.
+- [#2398](https://github.com/xmrig/xmrig/pull/2398) RandomX: optimized ARMv8 dataset read.
+- Added `argon2/ninja` alias for `argon2/wrkz` algorithm.
+
+# v6.12.1
+- [#2296](https://github.com/xmrig/xmrig/pull/2296) Fixed Zen3 assembly code for `cn/upx2` algorithm.
+
+# v6.12.0
+- [#2276](https://github.com/xmrig/xmrig/pull/2276) Added support for Uplexa (`cn/upx2` algorithm).
+- [#2261](https://github.com/xmrig/xmrig/pull/2261) Show total hashrate if compiled without OpenCL.
+- [#2289](https://github.com/xmrig/xmrig/pull/2289) RandomX: optimized `IMUL_RCP` instruction.
+- Added support for `--user` command line option for online benchmark.
+
+# v6.11.2
+- [#2207](https://github.com/xmrig/xmrig/issues/2207) Fixed regression in HTTP parser and llhttp updated to v5.1.0.
+
+# v6.11.1
+- [#2239](https://github.com/xmrig/xmrig/pull/2239) Fixed broken `coin` setting functionality.
+
+# v6.11.0
+- [#2196](https://github.com/xmrig/xmrig/pull/2196) Improved DNS subsystem and added new DNS specific options.
+- [#2172](https://github.com/xmrig/xmrig/pull/2172) Fixed build on Alpine 3.13.
+- [#2177](https://github.com/xmrig/xmrig/pull/2177) Fixed ARM specific compilation error with GCC 10.2.
+- [#2214](https://github.com/xmrig/xmrig/pull/2214) [#2216](https://github.com/xmrig/xmrig/pull/2216) [#2235](https://github.com/xmrig/xmrig/pull/2235) Optimized `cn-heavy` algorithm.
+- [#2217](https://github.com/xmrig/xmrig/pull/2217) Fixed mining job creation sequence.
+- [#2225](https://github.com/xmrig/xmrig/pull/2225) Fixed build without OpenCL support on some systems.
+- [#2229](https://github.com/xmrig/xmrig/pull/2229) Don't use RandomX JIT if `WITH_ASM=OFF`.
+- [#2228](https://github.com/xmrig/xmrig/pull/2228) Removed useless code for cryptonight algorithms.
+- [#2234](https://github.com/xmrig/xmrig/pull/2234) Fixed build error on gcc 4.8.
+
+# v6.10.0
+- [#2122](https://github.com/xmrig/xmrig/pull/2122) Fixed pause logic when both pause on battery and user activity are enabled.
+- [#2123](https://github.com/xmrig/xmrig/issues/2123) Fixed compatibility with gcc 4.8.
+- [#2147](https://github.com/xmrig/xmrig/pull/2147) Fixed many `new job` messages when solo mining.
+- [#2150](https://github.com/xmrig/xmrig/pull/2150) Updated `sse2neon.h` to the latest master, fixes build on ARMv7.
+- [#2157](https://github.com/xmrig/xmrig/pull/2157) Fixed crash in `cn-heavy` on Zen3 with manual thread count.
+- Fixed possible out of order write to log file.
+- [http-parser](https://github.com/nodejs/http-parser) replaced to [llhttp](https://github.com/nodejs/llhttp).
+- For official builds: libuv, hwloc and OpenSSL updated to latest versions.
+
+# v6.9.0
+- [#2104](https://github.com/xmrig/xmrig/pull/2104) Added [pause-on-active](https://xmrig.com/docs/miner/config/misc#pause-on-active) config option and `--pause-on-active=N` command line option.
+- [#2112](https://github.com/xmrig/xmrig/pull/2112) Added support for [Tari merge mining](https://github.com/tari-project/tari/blob/development/README.md#tari-merge-mining).
+- [#2117](https://github.com/xmrig/xmrig/pull/2117) Fixed crash when GPU mining `cn-heavy` on Zen3 system.
+
+# v6.8.2
+- [#2080](https://github.com/xmrig/xmrig/pull/2080) Fixed compile error in Termux.
+- [#2089](https://github.com/xmrig/xmrig/pull/2089) Optimized CryptoNight-Heavy for Zen3, 7-8% speedup.
+
+# v6.8.1
+- [#2064](https://github.com/xmrig/xmrig/pull/2064) Added documentation for config.json CPU options.
+- [#2066](https://github.com/xmrig/xmrig/issues/2066) Fixed AMD GPUs health data readings on Linux.
+- [#2067](https://github.com/xmrig/xmrig/pull/2067) Fixed compilation error when RandomX and Argon2 are disabled.
+- [#2076](https://github.com/xmrig/xmrig/pull/2076) Added support for flexible huge page sizes on Linux.
+- [#2077](https://github.com/xmrig/xmrig/pull/2077) Fixed `illegal instruction` crash on ARM.
+
+# v6.8.0
+- [#2052](https://github.com/xmrig/xmrig/pull/2052) Added DMI/SMBIOS reader.
+  - Added information about memory modules on the miner startup and for online benchmark.
+  - Added new HTTP API endpoint: `GET /2/dmi`.
+  - Added new command line option `--no-dmi` or config option `"dmi"`.
+  - Added new CMake option `-DWITH_DMI=OFF`.
+- [#2057](https://github.com/xmrig/xmrig/pull/2057) Improved MSR subsystem code quality.
+- [#2058](https://github.com/xmrig/xmrig/pull/2058) RandomX JIT x86: removed unnecessary instructions.
+
+# v6.7.2
+- [#2039](https://github.com/xmrig/xmrig/pull/2039) Fixed solo mining.
+
+# v6.7.1
+- [#1995](https://github.com/xmrig/xmrig/issues/1995) Fixed log initialization.
+- [#1998](https://github.com/xmrig/xmrig/pull/1998) Added hashrate in the benchmark finished message.
+- [#2009](https://github.com/xmrig/xmrig/pull/2009) AstroBWT OpenCL fixes.
+- [#2028](https://github.com/xmrig/xmrig/pull/2028) RandomX x86 JIT: removed redundant `CFROUND`.
+
 # v6.7.0
 - **[#1991](https://github.com/xmrig/xmrig/issues/1991) Added Apple M1 processor support.**
 - **[#1986](https://github.com/xmrig/xmrig/pull/1986) Up to 20-30% faster RandomX dataset initialization with AVX2 on some CPUs.**

CMakeLists.txt

@@ -1,14 +1,15 @@
-cmake_minimum_required(VERSION 2.8.12)
+cmake_minimum_required(VERSION 3.10)
 project(xmrig)
 
 option(WITH_HWLOC           "Enable hwloc support" ON)
 option(WITH_CN_LITE         "Enable CryptoNight-Lite algorithms family" ON)
 option(WITH_CN_HEAVY        "Enable CryptoNight-Heavy algorithms family" ON)
 option(WITH_CN_PICO         "Enable CryptoNight-Pico algorithm" ON)
+option(WITH_CN_FEMTO        "Enable CryptoNight-UPX2 algorithm" ON)
 option(WITH_RANDOMX         "Enable RandomX algorithms family" ON)
 option(WITH_ARGON2          "Enable Argon2 algorithms family" ON)
-option(WITH_ASTROBWT        "Enable AstroBWT algorithms family" ON)
 option(WITH_KAWPOW          "Enable KawPow algorithms family" ON)
+option(WITH_GHOSTRIDER      "Enable GhostRider algorithm" ON)
 option(WITH_HTTP            "Enable HTTP protocol support (client/server)" ON)
 option(WITH_DEBUG_LOG       "Enable debug log output" OFF)
 option(WITH_TLS             "Enable OpenSSL support" ON)
@@ -17,6 +18,8 @@ option(WITH_MSR             "Enable MSR mod & 1st-gen Ryzen fix" ON)
 option(WITH_ENV_VARS        "Enable environment variables support in config file" ON)
 option(WITH_EMBEDDED_CONFIG "Enable internal embedded JSON config" OFF)
 option(WITH_OPENCL          "Enable OpenCL backend" ON)
+set(WITH_OPENCL_VERSION 200 CACHE STRING "Target OpenCL version")
+set_property(CACHE WITH_OPENCL_VERSION PROPERTY STRINGS 120 200 210 220)
 option(WITH_CUDA            "Enable CUDA backend" ON)
 option(WITH_NVML            "Enable NVML (NVIDIA Management Library) support (only if CUDA backend enabled)" ON)
 option(WITH_ADL             "Enable ADL (AMD Display Library) or sysfs support (only if OpenCL backend enabled)" ON)
@@ -24,11 +27,15 @@ option(WITH_STRICT_CACHE    "Enable strict checks for OpenCL cache" ON)
 option(WITH_INTERLEAVE_DEBUG_LOG "Enable debug log for threads interleave" OFF)
 option(WITH_PROFILING       "Enable profiling for developers" OFF)
 option(WITH_SSE4_1          "Enable SSE 4.1 for Blake2" ON)
+option(WITH_AVX2            "Enable AVX2 for Blake2" ON)
+option(WITH_VAES            "Enable VAES instructions for Cryptonight" ON)
 option(WITH_BENCHMARK       "Enable builtin RandomX benchmark and stress test" ON)
 option(WITH_SECURE_JIT      "Enable secure access to JIT memory" OFF)
+option(WITH_DMI             "Enable DMI/SMBIOS reader" ON)
 
 option(BUILD_STATIC         "Build static binary" OFF)
-option(ARM_TARGET           "Force use specific ARM target 8 or 7" 0)
+option(ARM_V8               "Force ARMv8 (64 bit) architecture, use with caution if automatic detection fails, but you sure it may work" OFF)
+option(ARM_V7               "Force ARMv7 (32 bit) architecture, use with caution if automatic detection fails, but you sure it may work" OFF)
 option(HWLOC_DEBUG          "Enable hwloc debug helpers and log" OFF)
 
 
@@ -54,6 +61,7 @@ set(HEADERS
     src/core/config/usage.h
     src/core/Controller.h
     src/core/Miner.h
+    src/core/Taskbar.h
     src/net/interfaces/IJobResultListener.h
     src/net/JobResult.h
     src/net/JobResults.h
@@ -102,6 +110,7 @@ set(SOURCES
     src/core/config/ConfigTransform.cpp
     src/core/Controller.cpp
     src/core/Miner.cpp
+    src/core/Taskbar.cpp
     src/net/JobResults.cpp
     src/net/Network.cpp
     src/net/strategies/DonateStrategy.cpp
@@ -122,6 +131,19 @@ set(SOURCES_CRYPTO
     src/crypto/common/VirtualMemory.cpp
    )
 
+if (CMAKE_C_COMPILER_ID MATCHES GNU)
+    set_source_files_properties(src/crypto/cn/CnHash.cpp PROPERTIES COMPILE_FLAGS "-Ofast -fno-tree-vectorize")
+endif()
+
+if (WITH_VAES)
+    add_definitions(-DXMRIG_VAES)
+    set(HEADERS_CRYPTO "${HEADERS_CRYPTO}" src/crypto/cn/CryptoNight_x86_vaes.h)
+    set(SOURCES_CRYPTO "${SOURCES_CRYPTO}" src/crypto/cn/CryptoNight_x86_vaes.cpp)
+    if (CMAKE_C_COMPILER_ID MATCHES GNU OR CMAKE_C_COMPILER_ID MATCHES Clang)
+        set_source_files_properties(src/crypto/cn/CryptoNight_x86_vaes.cpp PROPERTIES COMPILE_FLAGS "-Ofast -fno-tree-vectorize -mavx2 -mvaes")
+    endif()
+endif()
+
 if (WITH_HWLOC)
     list(APPEND HEADERS_CRYPTO
         src/crypto/common/NUMAMemoryPool.h
@@ -140,14 +162,16 @@ if (XMRIG_OS_WIN)
         src/crypto/common/VirtualMemory_win.cpp
         )
 
-    set(EXTRA_LIBS ws2_32 psapi iphlpapi userenv)
+    set(EXTRA_LIBS ws2_32 psapi iphlpapi userenv dbghelp)
 elseif (XMRIG_OS_APPLE)
     list(APPEND SOURCES_OS
         src/App_unix.cpp
         src/crypto/common/VirtualMemory_unix.cpp
         )
+
     find_library(IOKIT_LIBRARY IOKit)
-    set(EXTRA_LIBS ${IOKIT_LIBRARY})
+    find_library(CORESERVICES_LIBRARY CoreServices)
+    set(EXTRA_LIBS ${IOKIT_LIBRARY} ${CORESERVICES_LIBRARY})
 else()
     list(APPEND SOURCES_OS
         src/App_unix.cpp
@@ -169,15 +193,15 @@ else()
 endif()
 
 add_definitions(-DXMRIG_MINER_PROJECT -DXMRIG_JSON_SINGLE_LINE_ARRAY)
-add_definitions(-D__STDC_FORMAT_MACROS -DUNICODE)
+add_definitions(-D__STDC_FORMAT_MACROS -DUNICODE -D_FILE_OFFSET_BITS=64)
 
 find_package(UV REQUIRED)
 
 include(cmake/flags.cmake)
 include(cmake/randomx.cmake)
 include(cmake/argon2.cmake)
-include(cmake/astrobwt.cmake)
 include(cmake/kawpow.cmake)
+include(cmake/ghostrider.cmake)
 include(cmake/OpenSSL.cmake)
 include(cmake/asm.cmake)
 
@@ -193,10 +217,17 @@ if (WITH_CN_PICO)
     add_definitions(/DXMRIG_ALGO_CN_PICO)
 endif()
 
+if (WITH_CN_FEMTO)
+    add_definitions(/DXMRIG_ALGO_CN_FEMTO)
+endif()
+
 if (WITH_EMBEDDED_CONFIG)
     add_definitions(/DXMRIG_FEATURE_EMBEDDED_CONFIG)
 endif()
 
+include(src/hw/api/api.cmake)
+include(src/hw/dmi/dmi.cmake)
+
 include_directories(src)
 include_directories(src/3rdparty)
 include_directories(${UV_INCLUDE_DIR})
@@ -206,16 +237,20 @@ if (WITH_DEBUG_LOG)
 endif()
 
 add_executable(${CMAKE_PROJECT_NAME} ${HEADERS} ${SOURCES} ${SOURCES_OS} ${HEADERS_CRYPTO} ${SOURCES_CRYPTO} ${SOURCES_SYSLOG} ${TLS_SOURCES} ${XMRIG_ASM_SOURCES})
-target_link_libraries(${CMAKE_PROJECT_NAME} ${XMRIG_ASM_LIBRARY} ${OPENSSL_LIBRARIES} ${UV_LIBRARIES} ${EXTRA_LIBS} ${CPUID_LIB} ${ARGON2_LIBRARY} ${ETHASH_LIBRARY})
+target_link_libraries(${CMAKE_PROJECT_NAME} ${XMRIG_ASM_LIBRARY} ${OPENSSL_LIBRARIES} ${UV_LIBRARIES} ${EXTRA_LIBS} ${CPUID_LIB} ${ARGON2_LIBRARY} ${ETHASH_LIBRARY} ${GHOSTRIDER_LIBRARY})
 
 if (WIN32)
-    add_custom_command(TARGET ${CMAKE_PROJECT_NAME} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different "${CMAKE_SOURCE_DIR}/bin/WinRing0/WinRing0x64.sys" $<TARGET_FILE_DIR:${CMAKE_PROJECT_NAME}>)
+    if (NOT ARM_TARGET)
+        add_custom_command(TARGET ${CMAKE_PROJECT_NAME} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different "${CMAKE_SOURCE_DIR}/bin/WinRing0/WinRing0x64.sys" $<TARGET_FILE_DIR:${CMAKE_PROJECT_NAME}>)
+    endif()
+
     add_custom_command(TARGET ${CMAKE_PROJECT_NAME} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different "${CMAKE_SOURCE_DIR}/scripts/benchmark_1M.cmd" $<TARGET_FILE_DIR:${CMAKE_PROJECT_NAME}>)
     add_custom_command(TARGET ${CMAKE_PROJECT_NAME} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different "${CMAKE_SOURCE_DIR}/scripts/benchmark_10M.cmd" $<TARGET_FILE_DIR:${CMAKE_PROJECT_NAME}>)
     add_custom_command(TARGET ${CMAKE_PROJECT_NAME} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different "${CMAKE_SOURCE_DIR}/scripts/pool_mine_example.cmd" $<TARGET_FILE_DIR:${CMAKE_PROJECT_NAME}>)
     add_custom_command(TARGET ${CMAKE_PROJECT_NAME} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different "${CMAKE_SOURCE_DIR}/scripts/solo_mine_example.cmd" $<TARGET_FILE_DIR:${CMAKE_PROJECT_NAME}>)
+    add_custom_command(TARGET ${CMAKE_PROJECT_NAME} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different "${CMAKE_SOURCE_DIR}/scripts/rtm_ghostrider_example.cmd" $<TARGET_FILE_DIR:${CMAKE_PROJECT_NAME}>)
 endif()
 
 if (CMAKE_CXX_COMPILER_ID MATCHES Clang AND CMAKE_BUILD_TYPE STREQUAL Release AND NOT CMAKE_GENERATOR STREQUAL Xcode)
-    add_custom_command(TARGET ${PROJECT_NAME} POST_BUILD COMMAND ${CMAKE_STRIP} ${CMAKE_PROJECT_NAME})
+    add_custom_command(TARGET ${PROJECT_NAME} POST_BUILD COMMAND ${CMAKE_STRIP} "$<TARGET_FILE:${CMAKE_PROJECT_NAME}>")
 endif()

README.md

@@ -7,10 +7,10 @@
 [![GitHub stars](https://img.shields.io/github/stars/xmrig/xmrig.svg)](https://github.com/xmrig/xmrig/stargazers)
 [![GitHub forks](https://img.shields.io/github/forks/xmrig/xmrig.svg)](https://github.com/xmrig/xmrig/network)
 
-XMRig is a high performance, open source, cross platform RandomX, KawPow, CryptoNight and AstroBWT unified CPU/GPU miner and [RandomX benchmark](https://xmrig.com/benchmark). Official binaries are available for Windows, Linux, macOS and FreeBSD.
+XMRig is a high performance, open source, cross platform RandomX, KawPow, CryptoNight and [GhostRider](https://github.com/xmrig/xmrig/tree/master/src/crypto/ghostrider#readme) unified CPU/GPU miner and [RandomX benchmark](https://xmrig.com/benchmark). Official binaries are available for Windows, Linux, macOS and FreeBSD.
 
 ## Mining backends
-- **CPU** (x64/ARMv8)
+- **CPU** (x86/x64/ARMv7/ARMv8)
 - **OpenCL** for AMD GPUs.
 - **CUDA** for NVIDIA GPUs via external [CUDA plugin](https://github.com/xmrig/xmrig-cuda).
 
@@ -19,7 +19,7 @@ XMRig is a high performance, open source, cross platform RandomX, KawPow, Crypto
 * **[Build from source](https://xmrig.com/docs/miner/build)**
 
 ## Usage
-The preferred way to configure the miner is the [JSON config file](src/config.json) as it is more flexible and human friendly. The [command line interface](https://xmrig.com/docs/miner/command-line-options) does not cover all features, such as mining profiles for different algorithms. Important options can be changed during runtime without miner restart by editing the config file or executing API calls.
+The preferred way to configure the miner is the [JSON config file](https://xmrig.com/docs/miner/config) as it is more flexible and human friendly. The [command line interface](https://xmrig.com/docs/miner/command-line-options) does not cover all features, such as mining profiles for different algorithms. Important options can be changed during runtime without miner restart by editing the config file or executing [API](https://xmrig.com/docs/miner/api) calls.
 
 * **[Wizard](https://xmrig.com/wizard)** helps you create initial configuration for the miner.
 * **[Workers](http://workers.xmrig.info)** helps manage your miners via HTTP API.

cmake/astrobwt.cmake

@@ -1,45 +0,0 @@
-if (WITH_ASTROBWT)
-    add_definitions(/DXMRIG_ALGO_ASTROBWT)
-
-    list(APPEND HEADERS_CRYPTO
-        src/crypto/astrobwt/AstroBWT.h
-    )
-
-    list(APPEND SOURCES_CRYPTO
-        src/crypto/astrobwt/AstroBWT.cpp
-    )
-
-    if (XMRIG_ARM)
-        list(APPEND HEADERS_CRYPTO
-            src/crypto/astrobwt/salsa20_ref/ecrypt-config.h
-            src/crypto/astrobwt/salsa20_ref/ecrypt-machine.h
-            src/crypto/astrobwt/salsa20_ref/ecrypt-portable.h
-            src/crypto/astrobwt/salsa20_ref/ecrypt-sync.h
-        )
-
-        list(APPEND SOURCES_CRYPTO
-            src/crypto/astrobwt/salsa20_ref/salsa20.c
-        )
-    else()
-        if (CMAKE_SIZEOF_VOID_P EQUAL 8)
-            add_definitions(/DASTROBWT_AVX2)
-            if (CMAKE_C_COMPILER_ID MATCHES MSVC)
-                enable_language(ASM_MASM)
-                list(APPEND SOURCES_CRYPTO src/crypto/astrobwt/sha3_256_avx2.asm)
-            else()
-                enable_language(ASM)
-                list(APPEND SOURCES_CRYPTO src/crypto/astrobwt/sha3_256_avx2.S)
-            endif()
-        endif()
-
-        list(APPEND HEADERS_CRYPTO
-            src/crypto/astrobwt/Salsa20.hpp
-        )
-
-        list(APPEND SOURCES_CRYPTO
-            src/crypto/astrobwt/Salsa20.cpp
-        )
-    endif()
-else()
-    remove_definitions(/DXMRIG_ALGO_ASTROBWT)
-endif()

cmake/cpu.cmake

@@ -1,47 +1,72 @@
+if (CMAKE_SIZEOF_VOID_P EQUAL 8)
+    set(XMRIG_64_BIT ON)
+    add_definitions(-DXMRIG_64_BIT)
+else()
+    set(XMRIG_64_BIT OFF)
+endif()
+
 if (NOT CMAKE_SYSTEM_PROCESSOR)
     message(WARNING "CMAKE_SYSTEM_PROCESSOR not defined")
 endif()
 
-if (CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64)$" AND CMAKE_SIZEOF_VOID_P EQUAL 8)
+include(CheckCXXCompilerFlag)
-    add_definitions(/DRAPIDJSON_SSE2)
+
+if (CMAKE_CXX_COMPILER_ID MATCHES MSVC)
+    set(VAES_SUPPORTED ON)
+else()
+    CHECK_CXX_COMPILER_FLAG("-mavx2 -mvaes" VAES_SUPPORTED)
+endif()
+
+if (NOT VAES_SUPPORTED)
+    set(WITH_VAES OFF)
+endif()
+
+if (XMRIG_64_BIT AND CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64)$")
+    add_definitions(-DRAPIDJSON_SSE2)
 else()
     set(WITH_SSE4_1 OFF)
+    set(WITH_AVX2 OFF)
+    set(WITH_VAES OFF)
+endif()
+
+add_definitions(-DRAPIDJSON_WRITE_DEFAULT_FLAGS=6) # rapidjson::kWriteNanAndInfFlag | rapidjson::kWriteNanAndInfNullFlag
+
+if (ARM_V8)
+    set(ARM_TARGET 8)
+elseif (ARM_V7)
+    set(ARM_TARGET 7)
 endif()
 
 if (NOT ARM_TARGET)
-    if (CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm64|armv8-a)$")
+    if (CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm64|ARM64|armv8-a)$")
         set(ARM_TARGET 8)
-    elseif (CMAKE_SYSTEM_PROCESSOR MATCHES "^(armv7|armv7f|armv7s|armv7k|armv7-a|armv7l)$")
+    elseif (CMAKE_SYSTEM_PROCESSOR MATCHES "^(armv7|armv7f|armv7s|armv7k|armv7-a|armv7l|armv7ve)$")
         set(ARM_TARGET 7)
     endif()
 endif()
 
 if (ARM_TARGET AND ARM_TARGET GREATER 6)
-    set(XMRIG_ARM     ON)
+    set(XMRIG_ARM ON)
-    add_definitions(/DXMRIG_ARM)
+    add_definitions(-DXMRIG_ARM=${ARM_TARGET})
 
     message(STATUS "Use ARM_TARGET=${ARM_TARGET} (${CMAKE_SYSTEM_PROCESSOR})")
 
-    include(CheckCXXCompilerFlag)
+    if (ARM_TARGET EQUAL 8 AND (CMAKE_CXX_COMPILER_ID MATCHES GNU OR CMAKE_CXX_COMPILER_ID MATCHES Clang))
-
-    if (ARM_TARGET EQUAL 8)
-        set(XMRIG_ARMv8 ON)
-        add_definitions(/DXMRIG_ARMv8)
-
         CHECK_CXX_COMPILER_FLAG(-march=armv8-a+crypto XMRIG_ARM_CRYPTO)
 
         if (XMRIG_ARM_CRYPTO)
-            add_definitions(/DXMRIG_ARM_CRYPTO)
+            add_definitions(-DXMRIG_ARM_CRYPTO)
             set(ARM8_CXX_FLAGS "-march=armv8-a+crypto")
         else()
             set(ARM8_CXX_FLAGS "-march=armv8-a")
         endif()
-    elseif (ARM_TARGET EQUAL 7)
-        set(XMRIG_ARMv7 ON)
-        add_definitions(/DXMRIG_ARMv7)
     endif()
 endif()
 
 if (WITH_SSE4_1)
-    add_definitions(/DXMRIG_FEATURE_SSE4_1)
+    add_definitions(-DXMRIG_FEATURE_SSE4_1)
+endif()
+
+if (WITH_AVX2)
+    add_definitions(-DXMRIG_FEATURE_AVX2)
 endif()

cmake/flags.cmake

@@ -10,7 +10,7 @@ if ("${CMAKE_BUILD_TYPE}" STREQUAL "")
 endif()
 
 if (CMAKE_BUILD_TYPE STREQUAL "Release")
-    add_definitions(/DNDEBUG)
+    add_definitions(-DNDEBUG)
 endif()
 
 include(CheckSymbolExists)
@@ -22,17 +22,17 @@ if (CMAKE_CXX_COMPILER_ID MATCHES GNU)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -fexceptions -fno-rtti -Wno-strict-aliasing -Wno-class-memaccess")
     set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -Ofast -s")
 
-    if (XMRIG_ARMv8)
+    if (ARM_TARGET EQUAL 8)
         set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${ARM8_CXX_FLAGS}")
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${ARM8_CXX_FLAGS} -flax-vector-conversions")
-    elseif (XMRIG_ARMv7)
+    elseif (ARM_TARGET EQUAL 7)
-        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mfpu=neon")
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -march=armv7-a -mfpu=neon")
-        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon -flax-vector-conversions")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=armv7-a -mfpu=neon -flax-vector-conversions")
     else()
         set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -maes")
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -maes")
 
-        add_definitions(/DHAVE_ROTR)
+        add_definitions(-DHAVE_ROTR)
     endif()
 
     if (WIN32)
@@ -49,41 +49,29 @@ if (CMAKE_CXX_COMPILER_ID MATCHES GNU)
         set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -static")
     endif()
 
-    add_definitions(/D_GNU_SOURCE)
+    add_definitions(-D_GNU_SOURCE -DHAVE_BUILTIN_CLEAR_CACHE)
-
-    if (${CMAKE_VERSION} VERSION_LESS "3.1.0")
-        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -std=c99")
-        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
-    endif()
-
-    #set(CMAKE_C_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -gdwarf-2")
-
-    add_definitions(/DHAVE_BUILTIN_CLEAR_CACHE)
 
 elseif (CMAKE_CXX_COMPILER_ID MATCHES MSVC)
-    set(CMAKE_C_FLAGS_RELEASE "/MT /O2 /Oi /DNDEBUG /GL")
+    set(CMAKE_C_FLAGS_RELEASE "/MP /MT /O2 /Oi /DNDEBUG /GL")
-    set(CMAKE_CXX_FLAGS_RELEASE "/MT /O2 /Oi /DNDEBUG /GL")
+    set(CMAKE_CXX_FLAGS_RELEASE "/MP /MT /O2 /Oi /DNDEBUG /GL")
 
-    set(CMAKE_C_FLAGS_RELWITHDEBINFO "/Ob1 /Zi /DRELWITHDEBINFO")
+    set(CMAKE_C_FLAGS_RELWITHDEBINFO "/MP /Ob1 /Zi /DRELWITHDEBINFO")
-    set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "/Ob1 /Zi /DRELWITHDEBINFO")
+    set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "/MP /Ob1 /Zi /DRELWITHDEBINFO")
 
-    add_definitions(/D_CRT_SECURE_NO_WARNINGS)
+    add_definitions(-D_CRT_SECURE_NO_WARNINGS -D_CRT_NONSTDC_NO_WARNINGS -DNOMINMAX -DHAVE_ROTR)
-    add_definitions(/D_CRT_NONSTDC_NO_WARNINGS)
-    add_definitions(/DNOMINMAX)
-    add_definitions(/DHAVE_ROTR)
 
 elseif (CMAKE_CXX_COMPILER_ID MATCHES Clang)
 
     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wall")
-    set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} -Ofast -funroll-loops -fmerge-all-constants")
+    set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} -funroll-loops -fmerge-all-constants")
 
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -fexceptions -fno-rtti -Wno-missing-braces")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -fexceptions -fno-rtti")
-    set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -Ofast -funroll-loops -fmerge-all-constants")
+    set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -funroll-loops -fmerge-all-constants")
 
-    if (XMRIG_ARMv8)
+    if (ARM_TARGET EQUAL 8)
         set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${ARM8_CXX_FLAGS}")
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${ARM8_CXX_FLAGS}")
-    elseif (XMRIG_ARMv7)
+    elseif (ARM_TARGET EQUAL 7)
         set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mfpu=neon -march=${CMAKE_SYSTEM_PROCESSOR}")
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon -march=${CMAKE_SYSTEM_PROCESSOR}")
     else()
@@ -92,19 +80,18 @@ elseif (CMAKE_CXX_COMPILER_ID MATCHES Clang)
 
         check_symbol_exists("_rotr" "x86intrin.h" HAVE_ROTR)
         if (HAVE_ROTR)
-            add_definitions(/DHAVE_ROTR)
+            add_definitions(-DHAVE_ROTR)
         endif()
     endif()
 
-    if (BUILD_STATIC)
+    if ((WIN32 AND ARM_TARGET) OR BUILD_STATIC)
         set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -static")
     endif()
-
 endif()
 
 if (NOT WIN32)
     check_symbol_exists("__builtin___clear_cache" "stdlib.h" HAVE_BUILTIN_CLEAR_CACHE)
     if (HAVE_BUILTIN_CLEAR_CACHE)
-        add_definitions(/DHAVE_BUILTIN_CLEAR_CACHE)
+        add_definitions(-DHAVE_BUILTIN_CLEAR_CACHE)
     endif()
 endif()

cmake/ghostrider.cmake

@@ -0,0 +1,8 @@
+if (WITH_GHOSTRIDER)
+    add_definitions(/DXMRIG_ALGO_GHOSTRIDER)
+    add_subdirectory(src/crypto/ghostrider)
+    set(GHOSTRIDER_LIBRARY ghostrider)
+else()
+    remove_definitions(/DXMRIG_ALGO_GHOSTRIDER)
+    set(GHOSTRIDER_LIBRARY "")
+endif()

cmake/os.cmake

@@ -15,39 +15,37 @@ else()
         set(XMRIG_OS_ANDROID ON)
     elseif(CMAKE_SYSTEM_NAME MATCHES "Linux")
         set(XMRIG_OS_LINUX ON)
-    elseif(CMAKE_SYSTEM_NAME STREQUAL FreeBSD)
+    elseif(CMAKE_SYSTEM_NAME STREQUAL FreeBSD OR CMAKE_SYSTEM_NAME STREQUAL DragonFly)
         set(XMRIG_OS_FREEBSD ON)
     endif()
 endif()
 
-
 if (XMRIG_OS_WIN)
-    add_definitions(/DWIN32)
+    add_definitions(-DWIN32 -DXMRIG_OS_WIN)
-    add_definitions(/DXMRIG_OS_WIN)
 elseif(XMRIG_OS_APPLE)
-    add_definitions(/DXMRIG_OS_APPLE)
+    add_definitions(-DXMRIG_OS_APPLE)
 
     if (XMRIG_OS_IOS)
-        add_definitions(/DXMRIG_OS_IOS)
+        add_definitions(-DXMRIG_OS_IOS)
     else()
-        add_definitions(/DXMRIG_OS_MACOS)
+        add_definitions(-DXMRIG_OS_MACOS)
     endif()
 
     if (XMRIG_ARM)
         set(WITH_SECURE_JIT ON)
     endif()
 elseif(XMRIG_OS_UNIX)
-    add_definitions(/DXMRIG_OS_UNIX)
+    add_definitions(-DXMRIG_OS_UNIX)
 
     if (XMRIG_OS_ANDROID)
-        add_definitions(/DXMRIG_OS_ANDROID)
+        add_definitions(-DXMRIG_OS_ANDROID)
     elseif (XMRIG_OS_LINUX)
-        add_definitions(/DXMRIG_OS_LINUX)
+        add_definitions(-DXMRIG_OS_LINUX)
     elseif (XMRIG_OS_FREEBSD)
-        add_definitions(/DXMRIG_OS_FREEBSD)
+        add_definitions(-DXMRIG_OS_FREEBSD)
     endif()
 endif()
 
 if (WITH_SECURE_JIT)
-    add_definitions(/DXMRIG_SECURE_JIT)
+    add_definitions(-DXMRIG_SECURE_JIT)
 endif()

cmake/randomx.cmake

@@ -1,4 +1,18 @@
 if (WITH_RANDOMX)
+    include(CheckSymbolExists)
+
+    if (WIN32)
+        check_symbol_exists(_aligned_malloc "stdlib.h" HAVE_ALIGNED_MALLOC)
+        if (HAVE_ALIGNED_MALLOC)
+            add_compile_definitions(HAVE_ALIGNED_MALLOC)
+        endif()
+    else()
+        check_symbol_exists(posix_memalign "stdlib.h" HAVE_POSIX_MEMALIGN)
+        if (HAVE_POSIX_MEMALIGN)
+            add_compile_definitions(HAVE_POSIX_MEMALIGN)
+        endif()
+    endif()
+
     add_definitions(/DXMRIG_ALGO_RANDOMX)
     set(WITH_ARGON2 ON)
 
@@ -42,13 +56,13 @@ if (WITH_RANDOMX)
         src/crypto/rx/RxVm.cpp
     )
 
-    if (CMAKE_C_COMPILER_ID MATCHES MSVC)
+    if (WITH_ASM AND CMAKE_C_COMPILER_ID MATCHES MSVC)
         enable_language(ASM_MASM)
         list(APPEND SOURCES_CRYPTO
              src/crypto/randomx/jit_compiler_x86_static.asm
              src/crypto/randomx/jit_compiler_x86.cpp
             )
-    elseif (NOT XMRIG_ARM AND CMAKE_SIZEOF_VOID_P EQUAL 8)
+    elseif (WITH_ASM AND NOT XMRIG_ARM AND CMAKE_SIZEOF_VOID_P EQUAL 8)
         list(APPEND SOURCES_CRYPTO
              src/crypto/randomx/jit_compiler_x86_static.S
              src/crypto/randomx/jit_compiler_x86.cpp
@@ -76,7 +90,15 @@ if (WITH_RANDOMX)
         list(APPEND SOURCES_CRYPTO src/crypto/randomx/blake2/blake2b_sse41.c)
 
         if (CMAKE_C_COMPILER_ID MATCHES GNU OR CMAKE_C_COMPILER_ID MATCHES Clang)
-            set_source_files_properties(src/crypto/randomx/blake2/blake2b_sse41.c PROPERTIES COMPILE_FLAGS -msse4.1)
+            set_source_files_properties(src/crypto/randomx/blake2/blake2b_sse41.c PROPERTIES COMPILE_FLAGS "-Ofast -msse4.1")
+        endif()
+    endif()
+
+    if (WITH_AVX2)
+        list(APPEND SOURCES_CRYPTO src/crypto/randomx/blake2/avx2/blake2b_avx2.c)
+
+        if (CMAKE_C_COMPILER_ID MATCHES GNU OR CMAKE_C_COMPILER_ID MATCHES Clang)
+            set_source_files_properties(src/crypto/randomx/blake2/avx2/blake2b_avx2.c PROPERTIES COMPILE_FLAGS "-Ofast -mavx2")
         endif()
     endif()
 
@@ -100,13 +122,29 @@ if (WITH_RANDOMX)
         message("-- WITH_MSR=ON")
 
         if (XMRIG_OS_WIN)
-            list(APPEND SOURCES_CRYPTO src/crypto/rx/Rx_win.cpp)
+            list(APPEND SOURCES_CRYPTO
+                src/crypto/rx/RxFix_win.cpp
+                src/hw/msr/Msr_win.cpp
+                )
         elseif (XMRIG_OS_LINUX)
-            list(APPEND SOURCES_CRYPTO src/crypto/rx/Rx_linux.cpp)
+            list(APPEND SOURCES_CRYPTO
+                src/crypto/rx/RxFix_linux.cpp
+                src/hw/msr/Msr_linux.cpp
+                )
         endif()
 
-        list(APPEND HEADERS_CRYPTO src/crypto/rx/msr/MsrItem.h)
+        list(APPEND HEADERS_CRYPTO
-        list(APPEND SOURCES_CRYPTO src/crypto/rx/msr/MsrItem.cpp)
+            src/crypto/rx/RxFix.h
+            src/crypto/rx/RxMsr.h
+            src/hw/msr/Msr.h
+            src/hw/msr/MsrItem.h
+            )
+
+        list(APPEND SOURCES_CRYPTO
+            src/crypto/rx/RxMsr.cpp
+            src/hw/msr/Msr.cpp
+            src/hw/msr/MsrItem.cpp
+            )
     else()
         remove_definitions(/DXMRIG_FEATURE_MSR)
         remove_definitions(/DXMRIG_FIX_RYZEN)

doc/ALGORITHMS.md

@@ -13,7 +13,6 @@ Option `coin` useful for pools without [algorithm negotiation](https://xmrig.com
 | Name | Memory | Version | Description | Notes |
 |------|--------|---------|-------------|-------|
 | `kawpow` | - | 6.0.0+ | KawPow (Ravencoin) | GPU only |
-| `rx/keva` | 1 MB | 5.9.0+ | RandomKEVA (RandomX variant for Keva). |  |
 | `astrobwt` | 20 MB | 5.8.0+ | AstroBWT (Dero). |  |
 | `cn-pico/tlo` | 256 KB | 5.5.0+ | CryptoNight-Pico (Talleo). |  |
 | `rx/sfx` | 2 MB | 5.4.0+ | RandomSFX (RandomX variant for Safex). |  |

doc/CHANGELOG_OLD.md

@@ -256,7 +256,7 @@
 
 # v2.8.0
 - **[#753](https://github.com/xmrig/xmrig/issues/753) Added new algorithm [CryptoNight variant 2](https://github.com/xmrig/xmrig/issues/753) for Monero fork, thanks [@SChernykh](https://github.com/SChernykh).**
-  - Added global and per thread option `"asm"` and and command line equivalent.
+  - Added global and per thread option `"asm"` and command line equivalent.
 - **[#758](https://github.com/xmrig/xmrig/issues/758) Added SSL/TLS support for secure connections to pools.**
   - Added per pool options `"tls"` and `"tls-fingerprint"` and command line equivalents.
 - [#767](https://github.com/xmrig/xmrig/issues/767) Added config autosave feature, same with GPU miners.  

doc/CPU.md

@@ -1,3 +1,5 @@
+**:warning: Recent version of this page https://xmrig.com/docs/miner/config/cpu.**
+
 # CPU backend
 
 All CPU related settings contains in one `cpu` object in config file, CPU backend allow specify multiple profiles and allow switch between them without restrictions by pool request or config change. Default auto-configuration create reasonable minimum of profiles which cover all supported algorithms.
@@ -75,6 +77,35 @@ Each number represent one thread and means CPU affinity, this is default format
 ```
 Internal format, but can be user defined.
 
+## RandomX options
+
+#### `init`
+Thread count to initialize RandomX dataset. Auto-detect (`-1`) or any number greater than 0 to use that many threads.
+
+#### `init-avx2`
+Use AVX2 for dataset initialization. Faster on some CPUs. Auto-detect (`-1`), disabled (`0`), always enabled on CPUs that support AVX2 (`1`).
+
+#### `mode`
+RandomX mining mode: `auto`, `fast` (2 GB memory), `light` (256 MB memory).
+
+#### `1gb-pages`
+Use 1GB hugepages for RandomX dataset (Linux only). Enabled (`true`) or disabled (`false`). It gives 1-3% speedup.
+
+#### `wrmsr`
+[MSR mod](https://xmrig.com/docs/miner/randomx-optimization-guide/msr). Enabled (`true`) or disabled (`false`). It gives up to 15% speedup depending on your system. _(**Note**: Userspace MSR writes are no longer enabled by default; the flag `msr.allow_writes=on` must be set for Linux Kernels 5.9 and after.)_
+
+#### `rdmsr`
+Restore MSR register values to their original values on exit. Used together with `wrmsr`. Enabled (`true`) or disabled (`false`).
+
+#### `cache_qos`
+[Cache QoS](https://xmrig.com/docs/miner/randomx-optimization-guide/qos). Enabled (`true`) or disabled (`false`). It's useful when you can't or don't want to mine on all CPU cores to make mining hashrate more stable.
+
+#### `numa`
+NUMA support (better hashrate on multi-CPU servers and Ryzen Threadripper 1xxx/2xxx). Enabled (`true`) or disabled (`false`).
+
+#### `scratchpad_prefetch_mode`
+Which instruction to use in RandomX loop to prefetch data from scratchpad. `1` is default and fastest in most cases. Can be off (`0`), `prefetcht0` instruction (`1`), `prefetchnta` instruction (`2`, a bit faster on Coffee Lake and a few other CPUs), `mov` instruction (`3`).
+
 ## Shared options
 
 #### `enabled`
@@ -83,23 +114,32 @@ Enable (`true`) or disable (`false`) CPU backend, by default `true`.
 #### `huge-pages`
 Enable (`true`) or disable (`false`) huge pages support, by default `true`.
 
+#### `huge-pages-jit`
+Enable (`true`) or disable (`false`) huge pages support for RandomX JIT code, by default `false`. It gives a very small boost on Ryzen CPUs, but hashrate is unstable between launches. Use with caution.
+
 #### `hw-aes`
 Force enable (`true`) or disable (`false`) hardware AES support. Default value `null` means miner autodetect this feature. Usually don't need change this option, this option useful for some rare cases when miner can't detect hardware AES, but it available. If you force enable this option, but your hardware not support it, miner will crash.
 
 #### `priority`
-Mining threads priority, value from `1` (lowest priority) to `5` (highest possible priority). Default value `null` means miner don't change threads priority at all.
+Mining threads priority, value from `1` (lowest priority) to `5` (highest possible priority). Default value `null` means miner don't change threads priority at all. Setting priority higher than 2 can make your PC unresponsive.
+
+#### `memory-pool` (since v4.3.0)
+Use continuous, persistent memory block for mining threads, useful for preserve huge pages allocation while algorithm switching. Possible values `false` (feature disabled, by default) or `true` or specific count of 2 MB huge pages. It helps to avoid loosing huge pages for scratchpads when RandomX dataset is updated and mining threads restart after a 2-3 days of mining.
+
+#### `yield` (since v5.1.1)
+Prefer system better system response/stability `true` (default value) or maximum hashrate `false`.
 
 #### `asm`
 Enable/configure or disable ASM optimizations. Possible values: `true`, `false`, `"intel"`, `"ryzen"`, `"bulldozer"`.
 
 #### `argon2-impl` (since v3.1.0)
-Allow override automatically detected Argon2 implementation, this option added mostly for debug purposes, default value `null` means autodetect. Other possible values: `"x86_64"`, `"SSE2"`, `"SSSE3"`, `"XOP"`, `"AVX2"`, `"AVX-512F"`. Manual selection has no safe guards, if you CPU not support required instuctions, miner will crash.
+Allow override automatically detected Argon2 implementation, this option added mostly for debug purposes, default value `null` means autodetect. This is used in RandomX dataset initialization and also in some other mining algorithms. Other possible values: `"x86_64"`, `"SSE2"`, `"SSSE3"`, `"XOP"`, `"AVX2"`, `"AVX-512F"`. Manual selection has no safe guards - if your CPU doesn't support required instuctions, miner will crash.
+
+#### `astrobwt-max-size`
+AstroBWT algorithm: skip hashes with large stage 2 size, default: `550`, min: `400`, max: `1200`. Optimal value depends on your CPU/GPU
+
+#### `astrobwt-avx2`
+AstroBWT algorithm: use AVX2 code. It's faster on some CPUs and slower on other
 
 #### `max-threads-hint` (since v4.2.0)
 Maximum CPU threads count (in percentage) hint for autoconfig. [CPU_MAX_USAGE.md](CPU_MAX_USAGE.md)
-
-#### `memory-pool` (since v4.3.0)
-Use continuous, persistent memory block for mining threads, useful for preserve huge pages allocation while algorithm swithing. Possible values `false` (feature disabled, by default) or `true` or specific count of 2 MB huge pages.
-
-#### `yield` (since v5.1.1)
-Prefer system better system response/stability `true` (default value) or maximum hashrate `false`.

doc/releases/5_0_1/SHA256SUMS

@@ -1,5 +0,0 @@
-6bb1a2e3a0fbca5195be6022f2a9fbff8a353c37c7542e7ab89420cb45b64505  xmrig-5.0.1-gcc-win32.zip
-24dba9ec281acfb2ea2c401ebd0e4e2d1f1ee5fd557da5ff3c7049020c1f78b6  xmrig-5.0.1-gcc-win64.zip
-86d65c6693ec9e35cd7547329580638b85c9eb0cf8383892a1c15199de5b556f  xmrig-5.0.1-msvc-cuda10_1-win64.zip
-0fbfe518b1c4b6993b0f66ff01302626375b15620ccf8f64d6fb97845068ffca  xmrig-5.0.1-msvc-win64.zip
-aa34890738a3494de2fa0e44db346937fea7339852f5f10b5d4655f95e2d8f1f  xmrig-5.0.1-xenial-x64.tar.gz

doc/releases/5_0_1/SHA256SUMS.sig

@@ -1,11 +0,0 @@
------BEGIN PGP SIGNATURE-----
-
-iQEzBAABCgAdFiEEmsTOqOZuNaXHzdwbRGpTY4vpRAkFAl3VcsoACgkQRGpTY4vp
-RAm9vQgA1MyTUU2jley2TCYLUzQy2Fffc8fbXYv64r44jbWOjC/6qo2iIlRgPhIc
-oVyPKr5TYS3QjDzCEm8IvozS0YudS6soESbPzqDonboK8pd0K4bsML9TQY2feV7A
-NL5vln0rfVHp1wxLLrQpfBqAgvJUXEyaHece6gFQN79JOGhEo2bHL2NyrOl+FViS
-b2BaMtXq410Fh+XT6ShnOaG/2EuO8ZqSGdCO6A/2LHQw1UY+mZiCvue6P6B06HmB
-WD/urOv38V389v+V+Sp4UlEW6VpBOOjvtChoVWtLt+tKzydrnt2EmoWWWg475pka
-4G6whHuMWS8CTt5/PDhJpvVXNQTIOw==
-=C764
------END PGP SIGNATURE-----

scripts/benchmark_10M.cmd

@@ -1,4 +1,4 @@
 @echo off
-cd %~dp0
+cd /d "%~dp0"
 xmrig.exe --bench=10M --submit
 pause

scripts/benchmark_1M.cmd

@@ -1,4 +1,4 @@
 @echo off
-cd %~dp0
+cd /d "%~dp0"
 xmrig.exe --bench=1M --submit
 pause

scripts/build.hwloc.sh

@@ -1,6 +1,10 @@
-#!/bin/bash -e
+#!/bin/sh -e
 
-HWLOC_VERSION="2.4.0"
+HWLOC_VERSION_MAJOR="2"
+HWLOC_VERSION_MINOR="12"
+HWLOC_VERSION_PATCH="1"
+
+HWLOC_VERSION="${HWLOC_VERSION_MAJOR}.${HWLOC_VERSION_MINOR}.${HWLOC_VERSION_PATCH}"
 
 mkdir -p deps
 mkdir -p deps/include
@@ -8,7 +12,7 @@ mkdir -p deps/lib
 
 mkdir -p build && cd build
 
-wget https://download.open-mpi.org/release/hwloc/v2.4/hwloc-${HWLOC_VERSION}.tar.gz -O hwloc-${HWLOC_VERSION}.tar.gz
+wget https://download.open-mpi.org/release/hwloc/v${HWLOC_VERSION_MAJOR}.${HWLOC_VERSION_MINOR}/hwloc-${HWLOC_VERSION}.tar.gz -O hwloc-${HWLOC_VERSION}.tar.gz
 tar -xzf hwloc-${HWLOC_VERSION}.tar.gz
 
 cd hwloc-${HWLOC_VERSION}
@@ -16,4 +20,4 @@ cd hwloc-${HWLOC_VERSION}
 make -j$(nproc || sysctl -n hw.ncpu || sysctl -n hw.logicalcpu)
 cp -fr include ../../deps
 cp hwloc/.libs/libhwloc.a ../../deps/lib
-cd ..
+cd ..

scripts/build.hwloc1.sh

@@ -1,4 +1,4 @@
-#!/bin/bash -e
+#!/bin/sh -e
 
 HWLOC_VERSION="1.11.13"
 

scripts/build.libressl.sh

@@ -1,6 +1,6 @@
-#!/bin/bash -e
+#!/bin/sh -e
 
-LIBRESSL_VERSION="3.0.2"
+LIBRESSL_VERSION="3.5.2"
 
 mkdir -p deps
 mkdir -p deps/include
@@ -17,4 +17,4 @@ make -j$(nproc || sysctl -n hw.ncpu || sysctl -n hw.logicalcpu)
 cp -fr include ../../deps
 cp crypto/.libs/libcrypto.a ../../deps/lib
 cp ssl/.libs/libssl.a ../../deps/lib
-cd ..
+cd ..

scripts/build.openssl.sh

@@ -1,6 +1,6 @@
-#!/bin/bash -e
+#!/bin/sh -e
 
-OPENSSL_VERSION="1.1.1i"
+OPENSSL_VERSION="1.1.1u"
 
 mkdir -p deps
 mkdir -p deps/include
@@ -8,7 +8,7 @@ mkdir -p deps/lib
 
 mkdir -p build && cd build
 
-wget https://www.openssl.org/source/openssl-${OPENSSL_VERSION}.tar.gz -O openssl-${OPENSSL_VERSION}.tar.gz
+wget https://openssl.org/source/old/1.1.1/openssl-${OPENSSL_VERSION}.tar.gz -O openssl-${OPENSSL_VERSION}.tar.gz
 tar -xzf openssl-${OPENSSL_VERSION}.tar.gz
 
 cd openssl-${OPENSSL_VERSION}
@@ -17,4 +17,4 @@ make -j$(nproc || sysctl -n hw.ncpu || sysctl -n hw.logicalcpu)
 cp -fr include ../../deps
 cp libcrypto.a ../../deps/lib
 cp libssl.a ../../deps/lib
-cd ..
+cd ..

scripts/build.openssl3.sh

@@ -0,0 +1,20 @@
+#!/bin/sh -e
+
+OPENSSL_VERSION="3.0.16"
+
+mkdir -p deps
+mkdir -p deps/include
+mkdir -p deps/lib
+
+mkdir -p build && cd build
+
+wget https://github.com/openssl/openssl/releases/download/openssl-${OPENSSL_VERSION}/openssl-${OPENSSL_VERSION}.tar.gz -O openssl-${OPENSSL_VERSION}.tar.gz
+tar -xzf openssl-${OPENSSL_VERSION}.tar.gz
+
+cd openssl-${OPENSSL_VERSION}
+./config -no-shared -no-asm -no-zlib -no-comp -no-dgram -no-filenames -no-cms
+make -j$(nproc || sysctl -n hw.ncpu || sysctl -n hw.logicalcpu)
+cp -fr include ../../deps
+cp libcrypto.a ../../deps/lib
+cp libssl.a ../../deps/lib
+cd ..

scripts/build.uv.sh

@@ -1,6 +1,6 @@
-#!/bin/bash -e
+#!/bin/sh -e
 
-UV_VERSION="1.40.0"
+UV_VERSION="1.51.0"
 
 mkdir -p deps
 mkdir -p deps/include
@@ -8,13 +8,13 @@ mkdir -p deps/lib
 
 mkdir -p build && cd build
 
-wget https://github.com/libuv/libuv/archive/v${UV_VERSION}.tar.gz -O v${UV_VERSION}.tar.gz
+wget https://dist.libuv.org/dist/v${UV_VERSION}/libuv-v${UV_VERSION}.tar.gz -O v${UV_VERSION}.tar.gz
 tar -xzf v${UV_VERSION}.tar.gz
 
-cd libuv-${UV_VERSION}
+cd libuv-v${UV_VERSION}
 sh autogen.sh
 ./configure --disable-shared
 make -j$(nproc || sysctl -n hw.ncpu || sysctl -n hw.logicalcpu)
 cp -fr include ../../deps
 cp .libs/libuv.a ../../deps/lib
-cd ..
+cd ..

scripts/build_deps.sh

@@ -1,5 +1,5 @@
-#!/bin/bash -e
+#!/bin/sh -e
 
 ./build.uv.sh
 ./build.hwloc.sh
-./build.openssl.sh
+./build.openssl3.sh

scripts/enable_1gb_pages.sh

@@ -1,4 +1,4 @@
-#!/bin/bash -e
+#!/bin/sh -e
 
 # https://xmrig.com/docs/miner/hugepages#onegb-huge-pages
 

scripts/generate_cl.js

@@ -6,7 +6,6 @@ const fs = require('fs');
 const path = require('path');
 const { text2h, text2h_bundle, addIncludes } = require('./js/opencl');
 const { opencl_minify } = require('./js/opencl_minify');
-const cwd = process.cwd();
 
 
 function cn()
@@ -50,7 +49,7 @@ function rx()
         'randomx_constants_monero.h',
         'randomx_constants_wow.h',
         'randomx_constants_arqma.h',
-        'randomx_constants_keva.h',
+        'randomx_constants_graft.h',
         'aes.cl',
         'blake2b.cl',
         'randomx_vm.cl',
@@ -66,15 +65,6 @@ function rx()
 }
 
 
-function astrobwt()
-{
-    const astrobwt = opencl_minify(addIncludes('astrobwt.cl', [ 'BWT.cl', 'salsa20.cl', 'sha3.cl' ]));
-
-    // fs.writeFileSync('astrobwt_gen.cl', astrobwt);
-    fs.writeFileSync('astrobwt_cl.h', text2h(astrobwt, 'xmrig', 'astrobwt_cl'));
-}
-
-
 function kawpow()
 {
     const kawpow = opencl_minify(addIncludes('kawpow.cl', [ 'defs.h' ]));
@@ -85,23 +75,24 @@ function kawpow()
     fs.writeFileSync('kawpow_dag_cl.h', text2h(kawpow_dag, 'xmrig', 'kawpow_dag_cl'));
 }
 
+for (let i = 0; i < 2; i++) {
+    if (fs.existsSync('src/backend/opencl/cl/OclSource.h')) {
+        break;
+    }
 
-process.chdir(path.resolve('src/backend/opencl/cl/cn'));
+    process.chdir('..');
+}
 
+process.chdir(path.resolve('src/backend/opencl/cl'));
+
+const cwd = process.cwd();
+
+process.chdir(path.resolve(cwd, 'cn'));
 cn();
 cn_r();
 
-process.chdir(cwd);
+process.chdir(path.resolve(cwd, 'rx'));
-process.chdir(path.resolve('src/backend/opencl/cl/rx'));
-
 rx();
 
-process.chdir(cwd);
+process.chdir(path.resolve(cwd, 'kawpow'));
-process.chdir(path.resolve('src/backend/opencl/cl/astrobwt'));
-
-astrobwt();
-
-process.chdir(cwd);
-process.chdir(path.resolve('src/backend/opencl/cl/kawpow'));
-
 kawpow();

scripts/pool_mine_example.cmd

@@ -15,6 +15,6 @@
 :: Choose pools outside of top 5 to help Monero network be more decentralized!
 :: Smaller pools also often have smaller fees/payout limits.
 
-cd %~dp0
+cd /d "%~dp0"
-xmrig.exe -o pool.hashvault.pro:3333 -u 48edfHu7V9Z84YzzMa6fUueoELZ9ZRXq9VetWzYGzKt52XU5xvqgzYnDK9URnRoJMk1j8nLwEVsaSWJ4fhdUyZijBGUicoD -p x
+xmrig.exe -o xmrpool.eu:3333 -u 48edfHu7V9Z84YzzMa6fUueoELZ9ZRXq9VetWzYGzKt52XU5xvqgzYnDK9URnRoJMk1j8nLwEVsaSWJ4fhdUyZijBGUicoD -p x
 pause

scripts/randomx_boost.sh

@@ -1,28 +1,52 @@
-#!/bin/bash
+#!/bin/sh -e
 
-modprobe msr
+MSR_FILE=/sys/module/msr/parameters/allow_writes
 
-if cat /proc/cpuinfo | grep "AMD Ryzen" > /dev/null;
+if test -e "$MSR_FILE"; then
+	echo on > $MSR_FILE
+else
+	modprobe msr allow_writes=on
+fi
+
+if grep -E 'AMD Ryzen|AMD EPYC|AuthenticAMD' /proc/cpuinfo > /dev/null;
 	then
-	if cat /proc/cpuinfo | grep "cpu family[[:space:]]:[[:space:]]25" > /dev/null;
+	if grep "cpu family[[:space:]]\{1,\}:[[:space:]]25" /proc/cpuinfo > /dev/null;
 		then
-			echo "Detected Ryzen (Zen3)"
+			if grep "model[[:space:]]\{1,\}:[[:space:]]97" /proc/cpuinfo > /dev/null;
-			wrmsr -a 0xc0011020 0x4480000000000
+				then
-			wrmsr -a 0xc0011021 0x1c000200000040
+					echo "Detected Zen4 CPU"
-			wrmsr -a 0xc0011022 0xc000000401500000
+					wrmsr -a 0xc0011020 0x4400000000000
-			wrmsr -a 0xc001102b 0x2000cc14
+					wrmsr -a 0xc0011021 0x4000000000040
-			echo "MSR register values for Ryzen (Zen3) applied"
+					wrmsr -a 0xc0011022 0x8680000401570000
+					wrmsr -a 0xc001102b 0x2040cc10
+					echo "MSR register values for Zen4 applied"
+				else
+					echo "Detected Zen3 CPU"
+					wrmsr -a 0xc0011020 0x4480000000000
+					wrmsr -a 0xc0011021 0x1c000200000040
+					wrmsr -a 0xc0011022 0xc000000401570000
+					wrmsr -a 0xc001102b 0x2000cc10
+					echo "MSR register values for Zen3 applied"
+				fi
+		elif grep "cpu family[[:space:]]\{1,\}:[[:space:]]26" /proc/cpuinfo > /dev/null;
+			then
+				echo "Detected Zen5 CPU"
+				wrmsr -a 0xc0011020 0x4400000000000
+				wrmsr -a 0xc0011021 0x4000000000040
+				wrmsr -a 0xc0011022 0x8680000401570000
+				wrmsr -a 0xc001102b 0x2040cc10
+				echo "MSR register values for Zen5 applied"
 		else
-			echo "Detected Ryzen (Zen1/Zen2)"
+			echo "Detected Zen1/Zen2 CPU"
 			wrmsr -a 0xc0011020 0
 			wrmsr -a 0xc0011021 0x40
 			wrmsr -a 0xc0011022 0x1510000
 			wrmsr -a 0xc001102b 0x2000cc16
-			echo "MSR register values for Ryzen (Zen1/Zen2) applied"
+			echo "MSR register values for Zen1/Zen2 applied"
 		fi
-elif cat /proc/cpuinfo | grep "Intel" > /dev/null;
+elif grep "Intel" /proc/cpuinfo > /dev/null;
 	then
-		echo "Detected Intel"
+		echo "Detected Intel CPU"
 		wrmsr -a 0x1a4 0xf
 		echo "MSR register values for Intel applied"
 else

scripts/rtm_ghostrider_example.cmd

@@ -0,0 +1,23 @@
+:: Example batch file for mining Raptoreum at a pool
+::
+:: Format:
+::      xmrig.exe -a gr -o <pool address>:<pool port> -u <pool username/wallet> -p <pool password>
+::
+:: Fields:
+::      pool address            The host name of the pool stratum or its IP address, for example raptoreumemporium.com
+::      pool port               The port of the pool's stratum to connect to, for example 3333. Check your pool's getting started page.
+::      pool username/wallet    For most pools, this is the wallet address you want to mine to. Some pools require a username
+::      pool password           For most pools this can be just 'x'. For pools using usernames, you may need to provide a password as configured on the pool.
+::
+:: List of Raptoreum mining pools:
+::      https://miningpoolstats.stream/raptoreum
+::
+:: Choose pools outside of top 5 to help Raptoreum network be more decentralized!
+:: Smaller pools also often have smaller fees/payout limits.
+
+cd /d "%~dp0"
+:: Use this command line to connect to non-SSL port
+xmrig.exe -a gr -o raptoreumemporium.com:3008 -u WALLET_ADDRESS -p x
+:: Or use this command line to connect to an SSL port
+:: xmrig.exe -a gr -o rtm.suprnova.cc:4273 --tls -u WALLET_ADDRESS -p x
+pause

scripts/solo_mine_example.cmd

@@ -11,6 +11,6 @@
 :: Mining solo is the best way to help Monero network be more decentralized!
 :: But you will only get a payout when you find a block which can take more than a year for a single low-end PC.
 
-cd %~dp0
+cd /d "%~dp0"
-xmrig.exe -o node.xmr.to:18081 -a rx/0 -u 48edfHu7V9Z84YzzMa6fUueoELZ9ZRXq9VetWzYGzKt52XU5xvqgzYnDK9URnRoJMk1j8nLwEVsaSWJ4fhdUyZijBGUicoD --daemon
+xmrig.exe -o YOUR_NODE_IP:18081 -a rx/0 -u 48edfHu7V9Z84YzzMa6fUueoELZ9ZRXq9VetWzYGzKt52XU5xvqgzYnDK9URnRoJMk1j8nLwEVsaSWJ4fhdUyZijBGUicoD --daemon
 pause

src/3rdparty/CL/cl_dx9_media_sharing.h

@@ -44,7 +44,7 @@ extern "C" {
 
 typedef cl_uint             cl_dx9_media_adapter_type_khr;
 typedef cl_uint             cl_dx9_media_adapter_set_khr;
-    
+
 #if defined(_WIN32)
 #include <d3d9.h>
 typedef struct _cl_dx9_surface_info_khr
@@ -105,7 +105,7 @@ typedef CL_API_ENTRY cl_mem (CL_API_CALL *clCreateFromDX9MediaSurfaceKHR_fn)(
     cl_mem_flags                  flags,
     cl_dx9_media_adapter_type_khr adapter_type,
     void *                        surface_info,
-    cl_uint                       plane,                                                                          
+    cl_uint                       plane,
     cl_int *                      errcode_ret) CL_API_SUFFIX__VERSION_1_2;
 
 typedef CL_API_ENTRY cl_int (CL_API_CALL *clEnqueueAcquireDX9MediaSurfacesKHR_fn)(

src/3rdparty/CL/cl_gl_ext.h

@@ -35,7 +35,7 @@ extern "C" {
 
 #include <CL/cl_gl.h>
 
-/* 
+/*
  *  cl_khr_gl_event extension
  */
 #define CL_COMMAND_GL_FENCE_SYNC_OBJECT_KHR     0x200D

src/3rdparty/adl/adl_defines.h

@@ -1471,7 +1471,7 @@ typedef enum _ADLProfilePropertyType
 #define ADL_HDR_FREESYNC_HDR    0x0004      ///< FreeSync HDR supported
 /// @}
 
-/// \defgroup define_FreesyncFlags ADLDDCInfo2 Freesync HDR flags 
+/// \defgroup define_FreesyncFlags ADLDDCInfo2 Freesync HDR flags
 /// @{
 /// defines for iFreesyncFlags in ADLDDCInfo2
 #define ADL_HDR_FREESYNC_BACKLIGHT_SUPPORT           0x0001      ///< Global backlight control supported
@@ -1738,7 +1738,7 @@ enum ADLODNDPMMaskType
      ADL_ODN_DPM_MASK                = 1 << 2,
 };
 
-//ODN features Bits for ADLODNCapabilitiesX2 
+//ODN features Bits for ADLODNCapabilitiesX2
 enum ADLODNFeatureControl
 {
      ADL_ODN_SCLK_DPM                = 1 << 0,
@@ -1764,7 +1764,7 @@ enum ADLODNFeatureControl
 
 //If any new feature is added, PPLIB only needs to add ext feature ID and Item ID(Seeting ID). These IDs should match the drive defined in CWDDEPM.h
 enum ADLODNExtFeatureControl
-{	
+{
 	ADL_ODN_EXT_FEATURE_MEMORY_TIMING_TUNE = 1 << 0,
 	ADL_ODN_EXT_FEATURE_FAN_ZERO_RPM_CONTROL = 1 << 1,
 	ADL_ODN_EXT_FEATURE_AUTO_UV_ENGINE = 1 << 2,   //Auto under voltage
@@ -1794,7 +1794,7 @@ enum ADLODNExtSettingId
 	ADL_ODN_PARAMETER_FAN_CURVE_SPEED_5,
     ADL_ODN_POWERGAUGE,
 	ODN_COUNT
-	
+
 } ;
 
 //OD8 Capability features bits
@@ -1811,7 +1811,7 @@ enum ADLOD8FeatureControl
     ADL_OD8_MEMORY_TIMING_TUNE = 1 << 8,
     ADL_OD8_FAN_ZERO_RPM_CONTROL = 1 << 9 ,
 	ADL_OD8_AUTO_UV_ENGINE = 1 << 10,  //Auto under voltage
-	ADL_OD8_AUTO_OC_ENGINE = 1 << 11,  //Auto overclock engine     
+	ADL_OD8_AUTO_OC_ENGINE = 1 << 11,  //Auto overclock engine
 	ADL_OD8_AUTO_OC_MEMORY = 1 << 12,  //Auto overclock memory
 	ADL_OD8_FAN_CURVE = 1 << 13,   //Fan curve
 	ADL_OD8_WS_AUTO_FAN_ACOUSTIC_LIMIT = 1 << 14, //Workstation Manual Fan controller
@@ -1888,7 +1888,7 @@ typedef enum _ADLSensorType
 	PMLOG_TEMPERATURE_VRSOC = 24,
 	PMLOG_TEMPERATURE_VRMVDD0 = 25,
 	PMLOG_TEMPERATURE_VRMVDD1 = 26,
-	PMLOG_TEMPERATURE_HOTSPOT = 27,    
+	PMLOG_TEMPERATURE_HOTSPOT = 27,
         PMLOG_TEMPERATURE_GFX = 28,
         PMLOG_TEMPERATURE_SOC = 29,
         PMLOG_GFX_POWER = 30,

src/3rdparty/adl/adl_sdk.h

@@ -37,7 +37,7 @@
 #define __stdcall
 #endif /* (LINUX) */
 
-/// Memory Allocation Call back 
+/// Memory Allocation Call back
 typedef void* ( __stdcall *ADL_MAIN_MALLOC_CALLBACK )( int );
 
 

src/3rdparty/adl/adl_structures.h

@@ -1753,7 +1753,7 @@ typedef struct ADLPXConfigCaps
 ///\brief Enum containing PX or HG type
 ///
 /// This enum is used to get PX or hG type
-/// 
+///
 /// \nosubgrouping
 //////////////////////////////////////////////////////////////////////////////////////////
 enum ADLPxType

src/3rdparty/argon2/CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 2.8.12)
+cmake_minimum_required(VERSION 3.10)
 
 project(argon2 C)
 set(CMAKE_C_STANDARD 99)

src/3rdparty/epee/LICENSE.txt

@@ -0,0 +1,25 @@
+Copyright (c) 2006-2013, Andrey N. Sabelnikov, www.sabelnikov.net
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+    * Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in the
+      documentation and/or other materials provided with the distribution.
+    * Neither the name of the Andrey N. Sabelnikov nor the
+      names of its contributors may be used to endorse or promote products
+      derived from this software without specific prior written permission.
+
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL Andrey N. Sabelnikov BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

src/3rdparty/epee/README.md

@@ -0,0 +1 @@
+epee -  is a small library of helpers, wrappers, tools and so on, used to make my life easier.

src/3rdparty/epee/span.h

@@ -0,0 +1,176 @@
+// Copyright (c) 2017-2020, The Monero Project
+//
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without modification, are
+// permitted provided that the following conditions are met:
+//
+// 1. Redistributions of source code must retain the above copyright notice, this list of
+//    conditions and the following disclaimer.
+//
+// 2. Redistributions in binary form must reproduce the above copyright notice, this list
+//    of conditions and the following disclaimer in the documentation and/or other
+//    materials provided with the distribution.
+//
+// 3. Neither the name of the copyright holder nor the names of its contributors may be
+//    used to endorse or promote products derived from this software without specific
+//    prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY
+// EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+// MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL
+// THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+// STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF
+// THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#pragma once
+
+#include <algorithm>
+#include <cstdint>
+#include <memory>
+#include <string>
+#include <type_traits>
+
+namespace epee
+{
+  /*!
+    \brief Non-owning sequence of data. Does not deep copy
+
+    Inspired by `gsl::span` and/or `boost::iterator_range`. This class is
+    intended to be used as a parameter type for functions that need to take a
+    writable or read-only sequence of data. Most common cases are `span<char>`
+    and `span<std::uint8_t>`. Using as a class member is only recommended if
+    clearly documented as not doing a deep-copy. C-arrays are easily convertible
+    to this type.
+
+    \note Conversion from C string literal to `span<const char>` will include
+      the NULL-terminator.
+    \note Never allows derived-to-base pointer conversion; an array of derived
+      types is not an array of base types.
+   */
+  template<typename T>
+  class span
+  {
+    template<typename U>
+    static constexpr bool safe_conversion() noexcept
+    {
+      // Allow exact matches or `T*` -> `const T*`.
+      using with_const = typename std::add_const<U>::type;
+      return std::is_same<T, U>() ||
+        (std::is_const<T>() && std::is_same<T, with_const>());
+    }
+
+  public:
+    using value_type = T;
+    using size_type = std::size_t;
+    using difference_type = std::ptrdiff_t;
+    using pointer = T*;
+    using const_pointer = const T*;
+    using reference = T&;
+    using const_reference = const T&;
+    using iterator = pointer;
+    using const_iterator = const_pointer;
+
+    constexpr span() noexcept : ptr(nullptr), len(0) {}
+    constexpr span(std::nullptr_t) noexcept : span() {}
+
+    //! Prevent derived-to-base conversions; invalid in this context.
+    template<typename U, typename = typename std::enable_if<safe_conversion<U>()>::type>
+    constexpr span(U* const src_ptr, const std::size_t count) noexcept
+      : ptr(src_ptr), len(count) {}
+
+    //! Conversion from C-array. Prevents common bugs with sizeof + arrays.
+    template<std::size_t N>
+    constexpr span(T (&src)[N]) noexcept : span(src, N) {}
+
+    constexpr span(const span&) noexcept = default;
+    span& operator=(const span&) noexcept = default;
+
+    /*! Try to remove `amount` elements from beginning of span.
+    \return Number of elements removed. */
+    std::size_t remove_prefix(std::size_t amount) noexcept
+    {
+        amount = std::min(len, amount);
+        ptr += amount;
+        len -= amount;
+        return amount;
+    }
+
+    constexpr iterator begin() const noexcept { return ptr; }
+    constexpr const_iterator cbegin() const noexcept { return ptr; }
+
+    constexpr iterator end() const noexcept { return begin() + size(); }
+    constexpr const_iterator cend() const noexcept { return cbegin() + size(); }
+
+    constexpr bool empty() const noexcept { return size() == 0; }
+    constexpr pointer data() const noexcept { return ptr; }
+    constexpr std::size_t size() const noexcept { return len; }
+    constexpr std::size_t size_bytes() const noexcept { return size() * sizeof(value_type); }
+
+    T &operator[](size_t idx) noexcept { return ptr[idx]; }
+    const T &operator[](size_t idx) const noexcept { return ptr[idx]; }
+
+  private:
+    T* ptr;
+    std::size_t len;
+  };
+
+  //! \return `span<const T::value_type>` from a STL compatible `src`.
+  template<typename T>
+  constexpr span<const typename T::value_type> to_span(const T& src)
+  {
+    // compiler provides diagnostic if size() is not size_t.
+    return {src.data(), src.size()};
+  }
+
+  //! \return `span<T::value_type>` from a STL compatible `src`.
+  template<typename T>
+  constexpr span<typename T::value_type> to_mut_span(T& src)
+  {
+    // compiler provides diagnostic if size() is not size_t.
+    return {src.data(), src.size()};
+  }
+
+  template<typename T>
+  constexpr bool has_padding() noexcept
+  {
+    return !std::is_standard_layout<T>() || alignof(T) != 1;
+  }
+
+  //! \return Cast data from `src` as `span<const std::uint8_t>`.
+  template<typename T>
+  span<const std::uint8_t> to_byte_span(const span<const T> src) noexcept
+  {
+    static_assert(!has_padding<T>(), "source type may have padding");
+    return {reinterpret_cast<const std::uint8_t*>(src.data()), src.size_bytes()};
+  }
+
+  //! \return `span<const std::uint8_t>` which represents the bytes at `&src`.
+  template<typename T>
+  span<const std::uint8_t> as_byte_span(const T& src) noexcept
+  {
+    static_assert(!std::is_empty<T>(), "empty types will not work -> sizeof == 1");
+    static_assert(!has_padding<T>(), "source type may have padding");
+    return {reinterpret_cast<const std::uint8_t*>(std::addressof(src)), sizeof(T)};
+  }
+
+  //! \return `span<std::uint8_t>` which represents the bytes at `&src`.
+  template<typename T>
+  span<std::uint8_t> as_mut_byte_span(T& src) noexcept
+  {
+    static_assert(!std::is_empty<T>(), "empty types will not work -> sizeof == 1");
+    static_assert(!has_padding<T>(), "source type may have padding");
+    return {reinterpret_cast<std::uint8_t*>(std::addressof(src)), sizeof(T)};
+  }
+
+  //! make a span from a std::string
+  template<typename T>
+  span<const T> strspan(const std::string &s) noexcept
+  {
+    static_assert(std::is_same<T, char>() || std::is_same<T, unsigned char>() || std::is_same<T, int8_t>() || std::is_same<T, uint8_t>(), "Unexpected type");
+    return {reinterpret_cast<const T*>(s.data()), s.size()};
+  }
+}

src/3rdparty/fmt/README.rst

@@ -81,7 +81,7 @@ Examples
 .. code:: c++
 
     #include <fmt/core.h>
-    
+
     int main() {
       fmt::print("Hello, world!\n");
     }
@@ -293,11 +293,11 @@ Projects using this library
   An open-source library for mathematical programming
 
 * `Aseprite <https://github.com/aseprite/aseprite>`_:
-  Animated sprite editor & pixel art tool 
+  Animated sprite editor & pixel art tool
 
 * `AvioBook <https://www.aviobook.aero/en>`_: A comprehensive aircraft
   operations suite
-  
+
 * `Celestia <https://celestia.space/>`_: Real-time 3D visualization of space
 
 * `Ceph <https://ceph.com/>`_: A scalable distributed storage system
@@ -351,7 +351,7 @@ Projects using this library
 
 * `quasardb <https://www.quasardb.net/>`_: A distributed, high-performance,
   associative database
-  
+
 * `Quill <https://github.com/odygrd/quill>`_: Asynchronous low-latency logging library
 
 * `QKW <https://github.com/ravijanjam/qkw>`_: Generalizing aliasing to simplify

src/3rdparty/getopt/getopt.h

@@ -3,9 +3,9 @@
  * DISCLAIMER
  * This file is part of the mingw-w64 runtime package.
  *
- * The mingw-w64 runtime package and its code is distributed in the hope that it 
+ * The mingw-w64 runtime package and its code is distributed in the hope that it
- * will be useful but WITHOUT ANY WARRANTY.  ALL WARRANTIES, EXPRESSED OR 
+ * will be useful but WITHOUT ANY WARRANTY.  ALL WARRANTIES, EXPRESSED OR
- * IMPLIED ARE HEREBY DISCLAIMED.  This includes but is not limited to 
+ * IMPLIED ARE HEREBY DISCLAIMED.  This includes but is not limited to
  * warranties of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  */
  /*
@@ -109,11 +109,7 @@ char    *optarg;		/* argument associated with option */
 extern char __declspec(dllimport) *__progname;
 #endif
 
-#ifdef __CYGWIN__
 static char EMSG[] = "";
-#else
-#define	EMSG		""
-#endif
 
 static int getopt_internal(int, char * const *, const char *,
 			   const struct option *, int *, int);

src/3rdparty/http-parser/AUTHORS

@@ -1,68 +0,0 @@
-# Authors ordered by first contribution.
-Ryan Dahl <ry@tinyclouds.org>
-Jeremy Hinegardner <jeremy@hinegardner.org>
-Sergey Shepelev <temotor@gmail.com>
-Joe Damato <ice799@gmail.com>
-tomika <tomika_nospam@freemail.hu>
-Phoenix Sol <phoenix@burninglabs.com>
-Cliff Frey <cliff@meraki.com>
-Ewen Cheslack-Postava <ewencp@cs.stanford.edu>
-Santiago Gala <sgala@apache.org>
-Tim Becker <tim.becker@syngenio.de>
-Jeff Terrace <jterrace@gmail.com>
-Ben Noordhuis <info@bnoordhuis.nl>
-Nathan Rajlich <nathan@tootallnate.net>
-Mark Nottingham <mnot@mnot.net>
-Aman Gupta <aman@tmm1.net>
-Tim Becker <tim.becker@kuriositaet.de>
-Sean Cunningham <sean.cunningham@mandiant.com>
-Peter Griess <pg@std.in>
-Salman Haq <salman.haq@asti-usa.com>
-Cliff Frey <clifffrey@gmail.com>
-Jon Kolb <jon@b0g.us>
-Fouad Mardini <f.mardini@gmail.com>
-Paul Querna <pquerna@apache.org>
-Felix Geisendörfer <felix@debuggable.com>
-koichik <koichik@improvement.jp>
-Andre Caron <andre.l.caron@gmail.com>
-Ivo Raisr <ivosh@ivosh.net>
-James McLaughlin <jamie@lacewing-project.org>
-David Gwynne <loki@animata.net>
-Thomas LE ROUX <thomas@november-eleven.fr>
-Randy Rizun <rrizun@ortivawireless.com>
-Andre Louis Caron <andre.louis.caron@usherbrooke.ca>
-Simon Zimmermann <simonz05@gmail.com>
-Erik Dubbelboer <erik@dubbelboer.com>
-Martell Malone <martellmalone@gmail.com>
-Bertrand Paquet <bpaquet@octo.com>
-BogDan Vatra <bogdan@kde.org>
-Peter Faiman <peter@thepicard.org>
-Corey Richardson <corey@octayn.net>
-Tóth Tamás <tomika_nospam@freemail.hu>
-Cam Swords <cam.swords@gmail.com>
-Chris Dickinson <christopher.s.dickinson@gmail.com>
-Uli Köhler <ukoehler@btronik.de>
-Charlie Somerville <charlie@charliesomerville.com>
-Patrik Stutz <patrik.stutz@gmail.com>
-Fedor Indutny <fedor.indutny@gmail.com>
-runner <runner.mei@gmail.com>
-Alexis Campailla <alexis@janeasystems.com>
-David Wragg <david@wragg.org>
-Vinnie Falco <vinnie.falco@gmail.com>
-Alex Butum <alexbutum@linux.com>
-Rex Feng <rexfeng@gmail.com>
-Alex Kocharin <alex@kocharin.ru>
-Mark Koopman <markmontymark@yahoo.com>
-Helge Heß <me@helgehess.eu>
-Alexis La Goutte <alexis.lagoutte@gmail.com>
-George Miroshnykov <george.miroshnykov@gmail.com>
-Maciej Małecki <me@mmalecki.com>
-Marc O'Morain <github.com@marcomorain.com>
-Jeff Pinner <jpinner@twitter.com>
-Timothy J Fontaine <tjfontaine@gmail.com>
-Akagi201 <akagi201@gmail.com>
-Romain Giraud <giraud.romain@gmail.com>
-Jay Satiro <raysatiro@yahoo.com>
-Arne Steen <Arne.Steen@gmx.de>
-Kjell Schubert <kjell.schubert@gmail.com>
-Olivier Mengué <dolmen@cpan.org>

src/3rdparty/http-parser/LICENSE-MIT

@@ -1,19 +0,0 @@
-Copyright Joyent, Inc. and other Node contributors.
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to
-deal in the Software without restriction, including without limitation the
-rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
-sell copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in
-all copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
-FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
-IN THE SOFTWARE. 

src/3rdparty/http-parser/README.md

@@ -1,246 +0,0 @@
-HTTP Parser
-===========
-
-[![Build Status](https://api.travis-ci.org/nodejs/http-parser.svg?branch=master)](https://travis-ci.org/nodejs/http-parser)
-
-This is a parser for HTTP messages written in C. It parses both requests and
-responses. The parser is designed to be used in performance HTTP
-applications. It does not make any syscalls nor allocations, it does not
-buffer data, it can be interrupted at anytime. Depending on your
-architecture, it only requires about 40 bytes of data per message
-stream (in a web server that is per connection).
-
-Features:
-
-  * No dependencies
-  * Handles persistent streams (keep-alive).
-  * Decodes chunked encoding.
-  * Upgrade support
-  * Defends against buffer overflow attacks.
-
-The parser extracts the following information from HTTP messages:
-
-  * Header fields and values
-  * Content-Length
-  * Request method
-  * Response status code
-  * Transfer-Encoding
-  * HTTP version
-  * Request URL
-  * Message body
-
-
-Usage
------
-
-One `http_parser` object is used per TCP connection. Initialize the struct
-using `http_parser_init()` and set the callbacks. That might look something
-like this for a request parser:
-```c
-http_parser_settings settings;
-settings.on_url = my_url_callback;
-settings.on_header_field = my_header_field_callback;
-/* ... */
-
-http_parser *parser = malloc(sizeof(http_parser));
-http_parser_init(parser, HTTP_REQUEST);
-parser->data = my_socket;
-```
-
-When data is received on the socket execute the parser and check for errors.
-
-```c
-size_t len = 80*1024, nparsed;
-char buf[len];
-ssize_t recved;
-
-recved = recv(fd, buf, len, 0);
-
-if (recved < 0) {
-  /* Handle error. */
-}
-
-/* Start up / continue the parser.
- * Note we pass recved==0 to signal that EOF has been received.
- */
-nparsed = http_parser_execute(parser, &settings, buf, recved);
-
-if (parser->upgrade) {
-  /* handle new protocol */
-} else if (nparsed != recved) {
-  /* Handle error. Usually just close the connection. */
-}
-```
-
-`http_parser` needs to know where the end of the stream is. For example, sometimes
-servers send responses without Content-Length and expect the client to
-consume input (for the body) until EOF. To tell `http_parser` about EOF, give
-`0` as the fourth parameter to `http_parser_execute()`. Callbacks and errors
-can still be encountered during an EOF, so one must still be prepared
-to receive them.
-
-Scalar valued message information such as `status_code`, `method`, and the
-HTTP version are stored in the parser structure. This data is only
-temporally stored in `http_parser` and gets reset on each new message. If
-this information is needed later, copy it out of the structure during the
-`headers_complete` callback.
-
-The parser decodes the transfer-encoding for both requests and responses
-transparently. That is, a chunked encoding is decoded before being sent to
-the on_body callback.
-
-
-The Special Problem of Upgrade
-------------------------------
-
-`http_parser` supports upgrading the connection to a different protocol. An
-increasingly common example of this is the WebSocket protocol which sends
-a request like
-
-        GET /demo HTTP/1.1
-        Upgrade: WebSocket
-        Connection: Upgrade
-        Host: example.com
-        Origin: http://example.com
-        WebSocket-Protocol: sample
-
-followed by non-HTTP data.
-
-(See [RFC6455](https://tools.ietf.org/html/rfc6455) for more information the
-WebSocket protocol.)
-
-To support this, the parser will treat this as a normal HTTP message without a
-body, issuing both on_headers_complete and on_message_complete callbacks. However
-http_parser_execute() will stop parsing at the end of the headers and return.
-
-The user is expected to check if `parser->upgrade` has been set to 1 after
-`http_parser_execute()` returns. Non-HTTP data begins at the buffer supplied
-offset by the return value of `http_parser_execute()`.
-
-
-Callbacks
----------
-
-During the `http_parser_execute()` call, the callbacks set in
-`http_parser_settings` will be executed. The parser maintains state and
-never looks behind, so buffering the data is not necessary. If you need to
-save certain data for later usage, you can do that from the callbacks.
-
-There are two types of callbacks:
-
-* notification `typedef int (*http_cb) (http_parser*);`
-    Callbacks: on_message_begin, on_headers_complete, on_message_complete.
-* data `typedef int (*http_data_cb) (http_parser*, const char *at, size_t length);`
-    Callbacks: (requests only) on_url,
-               (common) on_header_field, on_header_value, on_body;
-
-Callbacks must return 0 on success. Returning a non-zero value indicates
-error to the parser, making it exit immediately.
-
-For cases where it is necessary to pass local information to/from a callback,
-the `http_parser` object's `data` field can be used.
-An example of such a case is when using threads to handle a socket connection,
-parse a request, and then give a response over that socket. By instantiation
-of a thread-local struct containing relevant data (e.g. accepted socket,
-allocated memory for callbacks to write into, etc), a parser's callbacks are
-able to communicate data between the scope of the thread and the scope of the
-callback in a threadsafe manner. This allows `http_parser` to be used in
-multi-threaded contexts.
-
-Example:
-```c
- typedef struct {
-  socket_t sock;
-  void* buffer;
-  int buf_len;
- } custom_data_t;
-
-
-int my_url_callback(http_parser* parser, const char *at, size_t length) {
-  /* access to thread local custom_data_t struct.
-  Use this access save parsed data for later use into thread local
-  buffer, or communicate over socket
-  */
-  parser->data;
-  ...
-  return 0;
-}
-
-...
-
-void http_parser_thread(socket_t sock) {
- int nparsed = 0;
- /* allocate memory for user data */
- custom_data_t *my_data = malloc(sizeof(custom_data_t));
-
- /* some information for use by callbacks.
- * achieves thread -> callback information flow */
- my_data->sock = sock;
-
- /* instantiate a thread-local parser */
- http_parser *parser = malloc(sizeof(http_parser));
- http_parser_init(parser, HTTP_REQUEST); /* initialise parser */
- /* this custom data reference is accessible through the reference to the
- parser supplied to callback functions */
- parser->data = my_data;
-
- http_parser_settings settings; /* set up callbacks */
- settings.on_url = my_url_callback;
-
- /* execute parser */
- nparsed = http_parser_execute(parser, &settings, buf, recved);
-
- ...
- /* parsed information copied from callback.
- can now perform action on data copied into thread-local memory from callbacks.
- achieves callback -> thread information flow */
- my_data->buffer;
- ...
-}
-
-```
-
-In case you parse HTTP message in chunks (i.e. `read()` request line
-from socket, parse, read half headers, parse, etc) your data callbacks
-may be called more than once. `http_parser` guarantees that data pointer is only
-valid for the lifetime of callback. You can also `read()` into a heap allocated
-buffer to avoid copying memory around if this fits your application.
-
-Reading headers may be a tricky task if you read/parse headers partially.
-Basically, you need to remember whether last header callback was field or value
-and apply the following logic:
-
-    (on_header_field and on_header_value shortened to on_h_*)
-     ------------------------ ------------ --------------------------------------------
-    | State (prev. callback) | Callback   | Description/action                         |
-     ------------------------ ------------ --------------------------------------------
-    | nothing (first call)   | on_h_field | Allocate new buffer and copy callback data |
-    |                        |            | into it                                    |
-     ------------------------ ------------ --------------------------------------------
-    | value                  | on_h_field | New header started.                        |
-    |                        |            | Copy current name,value buffers to headers |
-    |                        |            | list and allocate new buffer for new name  |
-     ------------------------ ------------ --------------------------------------------
-    | field                  | on_h_field | Previous name continues. Reallocate name   |
-    |                        |            | buffer and append callback data to it      |
-     ------------------------ ------------ --------------------------------------------
-    | field                  | on_h_value | Value for current header started. Allocate |
-    |                        |            | new buffer and copy callback data to it    |
-     ------------------------ ------------ --------------------------------------------
-    | value                  | on_h_value | Value continues. Reallocate value buffer   |
-    |                        |            | and append callback data to it             |
-     ------------------------ ------------ --------------------------------------------
-
-
-Parsing URLs
-------------
-
-A simplistic zero-copy URL parser is provided as `http_parser_parse_url()`.
-Users of this library may wish to use it to parse URLs constructed from
-consecutive `on_url` callbacks.
-
-See examples of reading in headers:
-
-* [partial example](http://gist.github.com/155877) in C
-* [from http-parser tests](http://github.com/joyent/http-parser/blob/37a0ff8/test.c#L403) in C
-* [from Node library](http://github.com/joyent/node/blob/842eaf4/src/http.js#L284) in Javascript

src/3rdparty/http-parser/http_parser.h

@@ -1,442 +0,0 @@
-/* Copyright Joyent, Inc. and other Node contributors. All rights reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
- */
-#ifndef http_parser_h
-#define http_parser_h
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-/* Also update SONAME in the Makefile whenever you change these. */
-#define HTTP_PARSER_VERSION_MAJOR 2
-#define HTTP_PARSER_VERSION_MINOR 9
-#define HTTP_PARSER_VERSION_PATCH 3
-
-#include <stddef.h>
-#if defined(_WIN32) && !defined(__MINGW32__) && \
-  (!defined(_MSC_VER) || _MSC_VER<1600) && !defined(__WINE__)
-#include <BaseTsd.h>
-typedef __int8 int8_t;
-typedef unsigned __int8 uint8_t;
-typedef __int16 int16_t;
-typedef unsigned __int16 uint16_t;
-typedef __int32 int32_t;
-typedef unsigned __int32 uint32_t;
-typedef __int64 int64_t;
-typedef unsigned __int64 uint64_t;
-#else
-#include <stdint.h>
-#endif
-
-/* Compile with -DHTTP_PARSER_STRICT=0 to make less checks, but run
- * faster
- */
-#ifndef HTTP_PARSER_STRICT
-# define HTTP_PARSER_STRICT 1
-#endif
-
-/* Maximium header size allowed. If the macro is not defined
- * before including this header then the default is used. To
- * change the maximum header size, define the macro in the build
- * environment (e.g. -DHTTP_MAX_HEADER_SIZE=<value>). To remove
- * the effective limit on the size of the header, define the macro
- * to a very large number (e.g. -DHTTP_MAX_HEADER_SIZE=0x7fffffff)
- */
-#ifndef HTTP_MAX_HEADER_SIZE
-# define HTTP_MAX_HEADER_SIZE (80*1024)
-#endif
-
-typedef struct http_parser http_parser;
-typedef struct http_parser_settings http_parser_settings;
-
-
-/* Callbacks should return non-zero to indicate an error. The parser will
- * then halt execution.
- *
- * The one exception is on_headers_complete. In a HTTP_RESPONSE parser
- * returning '1' from on_headers_complete will tell the parser that it
- * should not expect a body. This is used when receiving a response to a
- * HEAD request which may contain 'Content-Length' or 'Transfer-Encoding:
- * chunked' headers that indicate the presence of a body.
- *
- * Returning `2` from on_headers_complete will tell parser that it should not
- * expect neither a body nor any futher responses on this connection. This is
- * useful for handling responses to a CONNECT request which may not contain
- * `Upgrade` or `Connection: upgrade` headers.
- *
- * http_data_cb does not return data chunks. It will be called arbitrarily
- * many times for each string. E.G. you might get 10 callbacks for "on_url"
- * each providing just a few characters more data.
- */
-typedef int (*http_data_cb) (http_parser*, const char *at, size_t length);
-typedef int (*http_cb) (http_parser*);
-
-
-/* Status Codes */
-#define HTTP_STATUS_MAP(XX)                                                 \
-  XX(100, CONTINUE,                        Continue)                        \
-  XX(101, SWITCHING_PROTOCOLS,             Switching Protocols)             \
-  XX(102, PROCESSING,                      Processing)                      \
-  XX(200, OK,                              OK)                              \
-  XX(201, CREATED,                         Created)                         \
-  XX(202, ACCEPTED,                        Accepted)                        \
-  XX(203, NON_AUTHORITATIVE_INFORMATION,   Non-Authoritative Information)   \
-  XX(204, NO_CONTENT,                      No Content)                      \
-  XX(205, RESET_CONTENT,                   Reset Content)                   \
-  XX(206, PARTIAL_CONTENT,                 Partial Content)                 \
-  XX(207, MULTI_STATUS,                    Multi-Status)                    \
-  XX(208, ALREADY_REPORTED,                Already Reported)                \
-  XX(226, IM_USED,                         IM Used)                         \
-  XX(300, MULTIPLE_CHOICES,                Multiple Choices)                \
-  XX(301, MOVED_PERMANENTLY,               Moved Permanently)               \
-  XX(302, FOUND,                           Found)                           \
-  XX(303, SEE_OTHER,                       See Other)                       \
-  XX(304, NOT_MODIFIED,                    Not Modified)                    \
-  XX(305, USE_PROXY,                       Use Proxy)                       \
-  XX(307, TEMPORARY_REDIRECT,              Temporary Redirect)              \
-  XX(308, PERMANENT_REDIRECT,              Permanent Redirect)              \
-  XX(400, BAD_REQUEST,                     Bad Request)                     \
-  XX(401, UNAUTHORIZED,                    Unauthorized)                    \
-  XX(402, PAYMENT_REQUIRED,                Payment Required)                \
-  XX(403, FORBIDDEN,                       Forbidden)                       \
-  XX(404, NOT_FOUND,                       Not Found)                       \
-  XX(405, METHOD_NOT_ALLOWED,              Method Not Allowed)              \
-  XX(406, NOT_ACCEPTABLE,                  Not Acceptable)                  \
-  XX(407, PROXY_AUTHENTICATION_REQUIRED,   Proxy Authentication Required)   \
-  XX(408, REQUEST_TIMEOUT,                 Request Timeout)                 \
-  XX(409, CONFLICT,                        Conflict)                        \
-  XX(410, GONE,                            Gone)                            \
-  XX(411, LENGTH_REQUIRED,                 Length Required)                 \
-  XX(412, PRECONDITION_FAILED,             Precondition Failed)             \
-  XX(413, PAYLOAD_TOO_LARGE,               Payload Too Large)               \
-  XX(414, URI_TOO_LONG,                    URI Too Long)                    \
-  XX(415, UNSUPPORTED_MEDIA_TYPE,          Unsupported Media Type)          \
-  XX(416, RANGE_NOT_SATISFIABLE,           Range Not Satisfiable)           \
-  XX(417, EXPECTATION_FAILED,              Expectation Failed)              \
-  XX(421, MISDIRECTED_REQUEST,             Misdirected Request)             \
-  XX(422, UNPROCESSABLE_ENTITY,            Unprocessable Entity)            \
-  XX(423, LOCKED,                          Locked)                          \
-  XX(424, FAILED_DEPENDENCY,               Failed Dependency)               \
-  XX(426, UPGRADE_REQUIRED,                Upgrade Required)                \
-  XX(428, PRECONDITION_REQUIRED,           Precondition Required)           \
-  XX(429, TOO_MANY_REQUESTS,               Too Many Requests)               \
-  XX(431, REQUEST_HEADER_FIELDS_TOO_LARGE, Request Header Fields Too Large) \
-  XX(451, UNAVAILABLE_FOR_LEGAL_REASONS,   Unavailable For Legal Reasons)   \
-  XX(500, INTERNAL_SERVER_ERROR,           Internal Server Error)           \
-  XX(501, NOT_IMPLEMENTED,                 Not Implemented)                 \
-  XX(502, BAD_GATEWAY,                     Bad Gateway)                     \
-  XX(503, SERVICE_UNAVAILABLE,             Service Unavailable)             \
-  XX(504, GATEWAY_TIMEOUT,                 Gateway Timeout)                 \
-  XX(505, HTTP_VERSION_NOT_SUPPORTED,      HTTP Version Not Supported)      \
-  XX(506, VARIANT_ALSO_NEGOTIATES,         Variant Also Negotiates)         \
-  XX(507, INSUFFICIENT_STORAGE,            Insufficient Storage)            \
-  XX(508, LOOP_DETECTED,                   Loop Detected)                   \
-  XX(510, NOT_EXTENDED,                    Not Extended)                    \
-  XX(511, NETWORK_AUTHENTICATION_REQUIRED, Network Authentication Required) \
-
-enum http_status
-  {
-#define XX(num, name, string) HTTP_STATUS_##name = num,
-  HTTP_STATUS_MAP(XX)
-#undef XX
-  };
-
-
-/* Request Methods */
-#define HTTP_METHOD_MAP(XX)         \
-  XX(0,  DELETE,      DELETE)       \
-  XX(1,  GET,         GET)          \
-  XX(2,  HEAD,        HEAD)         \
-  XX(3,  POST,        POST)         \
-  XX(4,  PUT,         PUT)          \
-  /* pathological */                \
-  XX(5,  CONNECT,     CONNECT)      \
-  XX(6,  OPTIONS,     OPTIONS)      \
-  XX(7,  TRACE,       TRACE)        \
-  /* WebDAV */                      \
-  XX(8,  COPY,        COPY)         \
-  XX(9,  LOCK,        LOCK)         \
-  XX(10, MKCOL,       MKCOL)        \
-  XX(11, MOVE,        MOVE)         \
-  XX(12, PROPFIND,    PROPFIND)     \
-  XX(13, PROPPATCH,   PROPPATCH)    \
-  XX(14, SEARCH,      SEARCH)       \
-  XX(15, UNLOCK,      UNLOCK)       \
-  XX(16, BIND,        BIND)         \
-  XX(17, REBIND,      REBIND)       \
-  XX(18, UNBIND,      UNBIND)       \
-  XX(19, ACL,         ACL)          \
-  /* subversion */                  \
-  XX(20, REPORT,      REPORT)       \
-  XX(21, MKACTIVITY,  MKACTIVITY)   \
-  XX(22, CHECKOUT,    CHECKOUT)     \
-  XX(23, MERGE,       MERGE)        \
-  /* upnp */                        \
-  XX(24, MSEARCH,     M-SEARCH)     \
-  XX(25, NOTIFY,      NOTIFY)       \
-  XX(26, SUBSCRIBE,   SUBSCRIBE)    \
-  XX(27, UNSUBSCRIBE, UNSUBSCRIBE)  \
-  /* RFC-5789 */                    \
-  XX(28, PATCH,       PATCH)        \
-  XX(29, PURGE,       PURGE)        \
-  /* CalDAV */                      \
-  XX(30, MKCALENDAR,  MKCALENDAR)   \
-  /* RFC-2068, section 19.6.1.2 */  \
-  XX(31, LINK,        LINK)         \
-  XX(32, UNLINK,      UNLINK)       \
-  /* icecast */                     \
-  XX(33, SOURCE,      SOURCE)       \
-
-enum http_method
-  {
-#define XX(num, name, string) HTTP_##name = num,
-  HTTP_METHOD_MAP(XX)
-#undef XX
-  };
-
-
-enum http_parser_type { HTTP_REQUEST, HTTP_RESPONSE, HTTP_BOTH };
-
-
-/* Flag values for http_parser.flags field */
-enum flags
-  { F_CHUNKED               = 1 << 0
-  , F_CONNECTION_KEEP_ALIVE = 1 << 1
-  , F_CONNECTION_CLOSE      = 1 << 2
-  , F_CONNECTION_UPGRADE    = 1 << 3
-  , F_TRAILING              = 1 << 4
-  , F_UPGRADE               = 1 << 5
-  , F_SKIPBODY              = 1 << 6
-  , F_CONTENTLENGTH         = 1 << 7
-  , F_TRANSFER_ENCODING     = 1 << 8
-  };
-
-
-/* Map for errno-related constants
- *
- * The provided argument should be a macro that takes 2 arguments.
- */
-#define HTTP_ERRNO_MAP(XX)                                           \
-  /* No error */                                                     \
-  XX(OK, "success")                                                  \
-                                                                     \
-  /* Callback-related errors */                                      \
-  XX(CB_message_begin, "the on_message_begin callback failed")       \
-  XX(CB_url, "the on_url callback failed")                           \
-  XX(CB_header_field, "the on_header_field callback failed")         \
-  XX(CB_header_value, "the on_header_value callback failed")         \
-  XX(CB_headers_complete, "the on_headers_complete callback failed") \
-  XX(CB_body, "the on_body callback failed")                         \
-  XX(CB_message_complete, "the on_message_complete callback failed") \
-  XX(CB_status, "the on_status callback failed")                     \
-  XX(CB_chunk_header, "the on_chunk_header callback failed")         \
-  XX(CB_chunk_complete, "the on_chunk_complete callback failed")     \
-                                                                     \
-  /* Parsing-related errors */                                       \
-  XX(INVALID_EOF_STATE, "stream ended at an unexpected time")        \
-  XX(HEADER_OVERFLOW,                                                \
-     "too many header bytes seen; overflow detected")                \
-  XX(CLOSED_CONNECTION,                                              \
-     "data received after completed connection: close message")      \
-  XX(INVALID_VERSION, "invalid HTTP version")                        \
-  XX(INVALID_STATUS, "invalid HTTP status code")                     \
-  XX(INVALID_METHOD, "invalid HTTP method")                          \
-  XX(INVALID_URL, "invalid URL")                                     \
-  XX(INVALID_HOST, "invalid host")                                   \
-  XX(INVALID_PORT, "invalid port")                                   \
-  XX(INVALID_PATH, "invalid path")                                   \
-  XX(INVALID_QUERY_STRING, "invalid query string")                   \
-  XX(INVALID_FRAGMENT, "invalid fragment")                           \
-  XX(LF_EXPECTED, "LF character expected")                           \
-  XX(INVALID_HEADER_TOKEN, "invalid character in header")            \
-  XX(INVALID_CONTENT_LENGTH,                                         \
-     "invalid character in content-length header")                   \
-  XX(UNEXPECTED_CONTENT_LENGTH,                                      \
-     "unexpected content-length header")                             \
-  XX(INVALID_CHUNK_SIZE,                                             \
-     "invalid character in chunk size header")                       \
-  XX(INVALID_TRANSFER_ENCODING,                                      \
-     "request has invalid transfer-encoding")                        \
-  XX(INVALID_CONSTANT, "invalid constant string")                    \
-  XX(INVALID_INTERNAL_STATE, "encountered unexpected internal state")\
-  XX(STRICT, "strict mode assertion failed")                         \
-  XX(PAUSED, "parser is paused")                                     \
-  XX(UNKNOWN, "an unknown error occurred")
-
-
-/* Define HPE_* values for each errno value above */
-#define HTTP_ERRNO_GEN(n, s) HPE_##n,
-enum http_errno {
-  HTTP_ERRNO_MAP(HTTP_ERRNO_GEN)
-};
-#undef HTTP_ERRNO_GEN
-
-
-/* Get an http_errno value from an http_parser */
-#define HTTP_PARSER_ERRNO(p)            ((enum http_errno) (p)->http_errno)
-
-
-struct http_parser {
-  /** PRIVATE **/
-  unsigned int type : 2;         /* enum http_parser_type */
-  unsigned int state : 7;        /* enum state from http_parser.c */
-  unsigned int header_state : 7; /* enum header_state from http_parser.c */
-  unsigned int index : 7;        /* index into current matcher */
-  unsigned int lenient_http_headers : 1;
-  unsigned int flags : 16;       /* F_* values from 'flags' enum; semi-public */
-
-  uint32_t nread;          /* # bytes read in various scenarios */
-  uint64_t content_length; /* # bytes in body (0 if no Content-Length header) */
-
-  /** READ-ONLY **/
-  unsigned short http_major;
-  unsigned short http_minor;
-  unsigned int status_code : 16; /* responses only */
-  unsigned int method : 8;       /* requests only */
-  unsigned int http_errno : 7;
-
-  /* 1 = Upgrade header was present and the parser has exited because of that.
-   * 0 = No upgrade header present.
-   * Should be checked when http_parser_execute() returns in addition to
-   * error checking.
-   */
-  unsigned int upgrade : 1;
-
-  /** PUBLIC **/
-  void *data; /* A pointer to get hook to the "connection" or "socket" object */
-};
-
-
-struct http_parser_settings {
-  http_cb      on_message_begin;
-  http_data_cb on_url;
-  http_data_cb on_status;
-  http_data_cb on_header_field;
-  http_data_cb on_header_value;
-  http_cb      on_headers_complete;
-  http_data_cb on_body;
-  http_cb      on_message_complete;
-  /* When on_chunk_header is called, the current chunk length is stored
-   * in parser->content_length.
-   */
-  http_cb      on_chunk_header;
-  http_cb      on_chunk_complete;
-};
-
-
-enum http_parser_url_fields
-  { UF_SCHEMA           = 0
-  , UF_HOST             = 1
-  , UF_PORT             = 2
-  , UF_PATH             = 3
-  , UF_QUERY            = 4
-  , UF_FRAGMENT         = 5
-  , UF_USERINFO         = 6
-  , UF_MAX              = 7
-  };
-
-
-/* Result structure for http_parser_parse_url().
- *
- * Callers should index into field_data[] with UF_* values iff field_set
- * has the relevant (1 << UF_*) bit set. As a courtesy to clients (and
- * because we probably have padding left over), we convert any port to
- * a uint16_t.
- */
-struct http_parser_url {
-  uint16_t field_set;           /* Bitmask of (1 << UF_*) values */
-  uint16_t port;                /* Converted UF_PORT string */
-
-  struct {
-    uint16_t off;               /* Offset into buffer in which field starts */
-    uint16_t len;               /* Length of run in buffer */
-  } field_data[UF_MAX];
-};
-
-
-/* Returns the library version. Bits 16-23 contain the major version number,
- * bits 8-15 the minor version number and bits 0-7 the patch level.
- * Usage example:
- *
- *   unsigned long version = http_parser_version();
- *   unsigned major = (version >> 16) & 255;
- *   unsigned minor = (version >> 8) & 255;
- *   unsigned patch = version & 255;
- *   printf("http_parser v%u.%u.%u\n", major, minor, patch);
- */
-unsigned long http_parser_version(void);
-
-void http_parser_init(http_parser *parser, enum http_parser_type type);
-
-
-/* Initialize http_parser_settings members to 0
- */
-void http_parser_settings_init(http_parser_settings *settings);
-
-
-/* Executes the parser. Returns number of parsed bytes. Sets
- * `parser->http_errno` on error. */
-size_t http_parser_execute(http_parser *parser,
-                           const http_parser_settings *settings,
-                           const char *data,
-                           size_t len);
-
-
-/* If http_should_keep_alive() in the on_headers_complete or
- * on_message_complete callback returns 0, then this should be
- * the last message on the connection.
- * If you are the server, respond with the "Connection: close" header.
- * If you are the client, close the connection.
- */
-int http_should_keep_alive(const http_parser *parser);
-
-/* Returns a string version of the HTTP method. */
-const char *http_method_str(enum http_method m);
-
-/* Returns a string version of the HTTP status code. */
-const char *http_status_str(enum http_status s);
-
-/* Return a string name of the given error */
-const char *http_errno_name(enum http_errno err);
-
-/* Return a string description of the given error */
-const char *http_errno_description(enum http_errno err);
-
-/* Initialize all http_parser_url members to 0 */
-void http_parser_url_init(struct http_parser_url *u);
-
-/* Parse a URL; return nonzero on failure */
-int http_parser_parse_url(const char *buf, size_t buflen,
-                          int is_connect,
-                          struct http_parser_url *u);
-
-/* Pause or un-pause the parser; a nonzero value pauses */
-void http_parser_pause(http_parser *parser, int paused);
-
-/* Checks if this is the final chunk of the body. */
-int http_body_is_final(const http_parser *parser);
-
-/* Change the maximum header size provided at compile time. */
-void http_parser_set_max_header_size(uint32_t size);
-
-#ifdef __cplusplus
-}
-#endif
-#endif

src/3rdparty/hwloc/CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required (VERSION  2.8.12)
+cmake_minimum_required(VERSION 3.10)
 project (hwloc C)
 
 include_directories(include)

src/3rdparty/hwloc/NEWS

@@ -1,5 +1,5 @@
 Copyright © 2009 CNRS
-Copyright © 2009-2020 Inria.  All rights reserved.
+Copyright © 2009-2025 Inria.  All rights reserved.
 Copyright © 2009-2013 Université Bordeaux
 Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
 Copyright © 2020 Hewlett Packard Enterprise.  All rights reserved.
@@ -17,6 +17,418 @@ bug fixes (and other actions) for each version of hwloc since version
 0.9.
 
 
+Version 2.12.1
+--------------
+* Add hwloc-calc's --default-nodes option to hwloc-bind and hwloc-info.
+* Improve the --best-memattr "default" fallback, try to use "default"
+  memory nodes, and add verbose messages and warnings if some
+  performance info are incomplete or missing.
+  Thanks to Antoine Morvan for the report.
+* Fix CPU and memory binding on different locations,
+  thanks to Antoine Morvan for the report.
+* Add HWLOC_LOCAL_NUMANODE_FLAG_INTERSECT_LOCALITY and enable it by
+  default in hwloc-calc --local-memory for finding local NUMA nodes
+  that do not exactly match input locations.
+  Thanks to Antoine Morvan for the report.
+* Fix a possible crash in the x86 backend when Qemu is configured to
+  expose multicore/thread CPUs that are actually single-core/thread.
+  Thanks to Georg Pfuetzenreuter.
+
+
+Version 2.12.0
+--------------
+* Add hwloc_topology_get_default_nodeset() for the set of default
+  NUMA nodes.
+  - hwloc-calc now has --default-nodes option.
+* Rework oneAPI LevelZero support to use zesInit() and avoid the need
+  to set ZES_ENABLE_SYSMAN=1 in the environment.
+  - zesDriverGetDeviceByUuidExp() is now required in the L0 runtime.
+  - ZES/Sysman variants were added in hwloc/levelzero.h to specifically
+    handle ZES/Sysman device handles.
+* Fix the locality of AMD GPU partitions, thanks to Edgar Leon for
+  reporting and debugging the issue.
+* Better detect Cray Slingshot NICs, thanks to Edgar Leon.
+* Add support for Die objects and Module groups on Windows.
+* Only filter-out Dies that are identical to their Packages
+  when it applies to all Dies.
+* Improve hwloc-calc to handle CPU-less NUMA nodes or platforms with
+  heterogeneous memory without requiring --nodeset-output.
+* hwloc-calc now accepts counting/listing cpukinds and memory tiers
+  with -N and -I cpukind/memorytier.
+* The systemd-dbus-api output of hwloc-calc has changed, and
+  --nodeset-output-format was added, to support NUMA node outputs.
+  Thanks to Pierre Neyron.
+* Update NVLink bandwidth and CUDA capabilities up to NVIDIA Blackwell.
+* Fix some NUMA syscalls on Linux for platforms with old libc headers.
+* Some minor fixes in distances.
+
+
+Version 2.11.2
+--------------
+* Add missing CPU info attrs on aarch64 on Linux.
+* Use ACPI CPPC on Linux to get better information about cpukinds,
+  at least on AMD CPUs.
+* Fix crash when manipulating cpukinds after topology
+  duplication, thanks to Hadrien Grasland for the report.
+* Fix missing input target checks in memattr functions,
+  thanks to Hadrien Grasland for the report.
+* Fix a memory leak when ignoring NUMA distances on FreeBSD.
+* Fix build failure on old Linux distributions without accessat().
+* Fix non-Windows importing of XML topologies and CPUID dumps exported
+  on Windows.
+* hwloc-calc --cpuset-output-format systemd-dbus-api now allows
+  to generate AllowedCPUs information for systemd slices.
+  See the hwloc-calc manpage for examples. Thanks to Pierre Neyron.
+* Some fixes in manpage EXAMPLES and split them into subsections.
+
+
+Version 2.11.1
+--------------
+* Fix bash completions, thanks Tavis Rudd.
+
+
+Version 2.11.0
+--------------
+* API
+  + Add HWLOC_MEMBIND_WEIGHTED_INTERLEAVE memory binding policy on
+    Linux 6.9+. Thanks to Honggyu Kim for the patch.
+    - weighted_interleave_membind is added to membind support bits.
+    - The "weighted" policy is added to the hwloc-bind tool.
+  + Add hwloc_obj_set_subtype(). Thanks to Hadrien Grasland for the report.
+* GPU support
+  + Don't hide the GPU NUMA node on NVIDIA Grace Hopper.
+  + Get Intel GPU OpenCL device locality.
+  + Add bandwidths between subdevices in the LevelZero XeLinkBandwidth
+    matrix.
+  + Fix PCI Gen4+ link speed of NVIDIA GPU obtained from NVML,
+    thanks to Akram Sbaih for the report.
+* Windows support
+  + Fix Windows support when UNICODE is enabled, several hwloc features
+    were missing, thanks to Martin for the report.
+  + Fix the enabling of CUDA in Windows CMake build,
+    Thanks to Moritz Kreutzer for the patch.
+  + Fix CUDA/OpenCL test source path in Windows CMake.
+* Tools
+  + Option --best-memattr may now return multiple nodes. Additional
+    configuration flags may be given to tweak its behavior.
+  + hwloc-info has a new --get-attr option to get a single attribute.
+  + hwloc-info now supports "levels", "support" and "topology"
+    special keywords for backward compatibility for hwloc 3.0.
+  + The --taskset command-line option is superseded by the new
+    --cpuset-output-format which also allows to export as list.
+  + hwloc-calc may now import bitmasks described as a list of bits
+    with the new "--cpuset-input-format list".
+* Misc
+  + The MemoryTiersNr info attribute in the root object now says how many
+    memory tiers were built. Thanks to Antoine Morvan for the report.
+  + Fix the management of infinite cpusets in the bitmap printf/sscanf
+    API as well as in command-line tools.
+  + Add section "Compiling software on top of hwloc's C API" in the
+    documentation with examples for GNU Make and CMake,
+    thanks to Florent Pruvost for the help.
+
+
+Version 2.10.0
+--------------
+* Heterogeneous Memory core improvements
+  + Better heuristics to identify the subtype of memory such as HBM,
+    DRAM, NVM, CXL-DRAM, etc.
+  + Build memory tiers, i.e. sets of NUMA nodes with the same subtype
+    and similar performance.
+    - NUMA node tier ranks are exposed in the new MemoryTier info
+      attribute (starts from 0 for highest bandwidth tier)..
+  + See the new Heterogeneous Memory section in the documentation.
+* API
+  + Add hwloc_topology_free_group_object() to discard a Group created
+    by hwloc_topology_alloc_group_object().
+* Linux backend
+  + Fix cpukinds on NVIDIA Grace to report identical cores even if they
+    actually have very small frequency differences.
+    Thanks to John C. Linford for the report.
+  + Add CXLDevice attributes to CXL DAX objects and NUMA nodes to show
+    which PCI device implements which window.
+  + Ignore buggy memory-side caches and memory attributes when fake NUMA
+    emulation is enabled on the Linux kernel command-line.
+  + Add more info attributes in MemoryModule Misc objects,
+    thanks to Zubiao Xiong for the patch.
+  + Get CPUModel and CPUFamily info attributes on LoongArch platforms.
+* x86 backend
+  + Add support for new AMD CPUID leaf 0x80000026 for better detection
+    of Core Complex and Die on Zen4 processors.
+  + Improve Zhaoxin CPU topology detection.
+* Tools
+  + Input locations and many command-line options (e.g. hwloc-calc -I -N -H,
+    lstopo --only) now accept filters such as "NUMA[HBM]" so that only
+    objects are that type and subtype are considered.
+    - NUMA[tier=1] is also accepted for selecting NUMA nodes depending
+      on their MemoryTier info attribute.
+  + Add --object-output to hwloc-calc to report the type as a prefix to
+    object indexes, e.g. Core:2 instead of 2 in the output of -I.
+  + hwloc-info --ancestor and --descendants now accepts kinds of objects
+    instead of single types.
+    - The new --first option only shows the first matching object.
+  + Add --children-of-pid to hwloc-ps to show a hierarchy of processes.
+    Thanks to Antoine Morvan for the suggestion.
+  + Add --misc-from to lstopo to add Misc objects described in a file.
+    - To be combined with the new hwloc-ps --lstopo-misc for a customizable
+      lstopo --top replacement.
+* Misc
+  + lstopo may now configure the layout of memory object placed above,
+    for instance with --children-order memory:above:vert.
+  + Fix XML import from memory or stdin when using libxml2 2.12.
+  + Fix installation failures when configuring with --target,
+    thanks to Clement Foyer for the patch.
+  + Fix support for 128bit pointer architectures.
+  + Remove Netloc.
+
+
+Version 2.9.3
+-------------
+* Handle Linux glibc allocation errors in binding routines (CVE-2022-47022).
+* Fix hwloc-calc when searching objects on heterogeneous memory platforms,
+  thanks to Antoine Morvan for the report.
+* Fix hwloc_get_next_child() when there are some memory-side caches.
+* Don't crash if the topology is empty because Linux cgroups are wrong.
+* Improve some hwloc-bind warnings in case of command-line parsing errors.
+* Many documentation improvements all over the place, including:
+  + hwloc_topology_restrict() and hwloc_topology_insert_group() may reorder
+    children, causing the logical indexes of objects to change.
+
+
+Version 2.9.2
+-------------
+* Don't forget L3i when defining filters for multiple levels of caches
+  with hwloc_topology_set_cache/icache_types_filter().
+* Fix object total_memory after hwloc_topology_insert_group_object().
+* Fix the (non-yet) exporting in synthetic description for complex memory
+  hierarchies with memory-side caches, etc.
+* Fix some default size attributes when building synthetic topologies.
+* Fix size units in hwloc-annotate.
+* Improve bitmap reallocation error management in many functions.
+* Documentation improvements:
+  + Better document return values of functions.
+  + Add "Error reporting" section (in hwloc.h and in the doxygen doc).
+  + Add FAQ entry "What may I disable to make hwloc faster?"
+  + Improve FAQ entries "Why is lstopo slow?" and
+    "I only need ..., why should I use hwloc?"
+  + Clarify how to deal with cpukinds in hwloc-calc and hwloc-bind
+    manpages.
+
+
+Version 2.9.1
+-------------
+* Don't forget to apply object type filters to "perflevel" caches detected
+  on recent Mac OS X releases, thanks to Michel Lesoinne for the report.
+* Fix a failed assertion in hwloc_topology_restrict() when some NUMA nodes
+  are removed because of HWLOC_RESTRICT_FLAG_REMOVE_CPULESS but no PUs are.
+  Thanks to Mark Grondona for reporting the issue.
+* Mark HPE Cray Slingshot NICs with subtype "Slingshot".
+
+
+Version 2.9.0
+-------------
+* Backends
+  + Expose the memory size of CXL memory devices (Type 3) on Linux.
+  + The LevelZero backend now reports the "XeLinkBandwidth" distance
+    matrix between L0 devices (and subdevices) when available.
+  + Add support for CUDA compute capability up to 9.0.
+* Tools
+  + lstopo now switches to console mode when its output is redirected.
+    Graphical window mode may be forced back with --of window.
+  + hwloc-calc now accepts "numa" in -H, and I/O subtypes such as "gpu"
+    in -I and -N.
+
+
+Version 2.8.0
+-------------
+* API
+  + Add HWLOC_TOPOLOGY_FLAG_NO_DISTANCES, _NO_MEMATTRS and _NO_CPUKINDS
+    to reduce the overhead when unneeded.
+  + Add separate Read/Write Bandwidth/Latency memory attributes and
+    implement them on Linux.
+* Backends
+  + NUMA nodes may now have a subtype such as DRAM, HBM, SPM, or NVM
+    on heterogeneous memory platforms on Linux.
+    - Add DAXType and DAXParent attributes on Linux to tell where a
+      DAX device or its corresponding NUMA node come from (SPM for
+      Specific-Purpose or NVM for Non-Volatile Memory).
+  + Detect heterogeneous caches in hybrid CPUs on MacOS X,
+    thanks to Paul Bone for the help.
+  + Max frequencies are not ignored in Linux cpukinds anymore (they were
+    ignored in hwloc 2.7.0), but they may be slightly adjusted to avoid
+    reporting hybrid CPUs because Intel Turbo Boost Max 3.0.
+    - See the documentation of environment variable HWLOC_CPUKINDS_MAXFREQ.
+  + Hardwire the PCI locality of HPE Cray EX235a nodes.
+* Tools
+  + lstopo and other tools may now load Linux and x86 cpuid topology files
+    from a tarball.
+  + lstopo may now replace the P# and L# index prefixes with custom strings
+    thanks to --os-index-prefix and --logical-index-prefix options.
+* Misc
+  + Add --disable-readme to avoid regenerating the top-level hwloc README
+    file from the documentation.
+
+
+Version 2.7.2
+-------------
+* Fix a crash when LevelZero devices have multiple subdevices,
+  e.g. on PonteVecchio GPUs, thanks to Jonathan Peyton.
+* Fix a leak when importing cpukinds from XML,
+  thanks to Hui Zhou.
+
+
+Version 2.7.1
+-------------
+* Workaround crashes when virtual machines report incoherent x86 CPUID
+  information about numbers of cores and threads.
+  Thanks to Peter Bense for the report.
+* Use setenv() instead of putenv() when trying to force enable oneAPI L0
+  support, to avoid issues with applications that touch the environment,
+  thanks to Josh Hursey for the patch.
+* Add some warnings at the end of configure when GPU libraries are
+  missing on the system or their path is missing in the environment.
+
+
+Version 2.7.0
+-------------
+* Backends
+  + Add support for NUMA nodes and caches with more than 64 PUs across
+    multiple processor groups on Windows 11 and Windows Server 2022.
+  + Group objects are not created for Windows processor groups anymore,
+    except if HWLOC_WINDOWS_PROCESSOR_GROUP_OBJS=1 in the environment.
+  + Expose "Cluster" group objects on Linux kernel 5.16+ for CPUs
+    that share some internal cache or bus. This can be equivalent
+    to the L2 Cache level on some platforms (e.g. x86) or a specific
+    level between L2 and L3 on others (e.g. ARM Kungpeng 920).
+    Thanks to Jonathan Cameron for the help.
+    - HWLOC_DONT_MERGE_CLUSTER_GROUPS=1 may be set in the environment
+      to prevent these groups from being merged with identical caches, etc.
+  + Improve the oneAPI LevelZero backend:
+    - Expose subdevices such as "ze0.1" inside root OS devices ("ze0")
+      when the hardware contains multiple subdevices.
+    - Add many new attributes to describe device type, and the
+      numbers of slices, subslices, execution units and threads.
+    - Expose the memory information as LevelZeroHBM/DDR/MemorySize infos.
+  + Ignore the max frequencies of cores in Linux cpukinds when the
+    base frequencies are available (to avoid exposing hybrid CPUs
+    when Intel Turbo Boost Max 3.0 gives slightly different max
+    frequencies to CPU cores).
+    - May be reverted by setting HWLOC_CPUKINDS_MAXFREQ=1 in the environment.
+* Tools
+  + Add --grey and --palette options to switch lstopo to greyscale or
+    white-background-only graphics, or to tune individual colors.
+* Build
+  + Windows CMake builds now support non-MSVC compilers, detect several
+    features at build time, can build/run tests, etc.
+    Thanks to Michael Hirsch and Alexander Neumann .
+
+
+Version 2.6.0
+-------------
+* Backends
+  + Expose two cpukinds for energy-efficient cores (icestorm) and
+    high-performance cores (firestorm) on Apple M1 on Mac OS X.
+  + Use sysfs CPU "capacity" to rank hybrid cores by efficiency
+    on Linux when available (mostly on recent ARM platforms for now).
+  + Improve HWLOC_MEMBIND_BIND (without the STRICT flag) on Linux kernel
+    >= 5.15: If more than one node is given, the kernel may now use all
+    of them instead of only the first one before falling back to others.
+  + Expose cache os_index when available on Linux, it may be needed
+    when using resctrl to configure cache partitioning, memory bandwidth
+    monitoring, etc.
+  + Add a "XGMIHops" distances matrix in the RSMI backend for AMD GPU
+    interconnected through XGMI links.
+  + Expose AMD GPU memory information (VRAM and GTT) in the RSMI backend.
+  + Add OS devices such as "bxi0" for Atos/Bull BXI HCAs on Linux.
+* Tools
+  + lstopo has a better placement algorithm with respect to I/O
+    objects, see --children-order in the manpage for details.
+  + hwloc-annotate may now change object subtypes and cache or memory
+    sizes.
+* Build
+  + Allow to specify the ROCm installation for building the RSMI backend:
+    - Use a custom installation path if specified with --with-rocm=<dir>.
+    - Use /opt/rocm-<version> if specified with --with-rocm-version=<version>
+      or the ROCM_VERSION environment variable.
+    - Try /opt/rocm if it exists.
+    - See "How do I enable ROCm SMI and select which version to use?"
+      in the FAQ for details.
+  + Add a CMakeLists for Windows under contrib/windows-cmake/ .
+* Documentation
+  + Add FAQ entry "How do I create a custom heterogeneous and
+     asymmetric topology?"
+
+
+Version 2.5.0
+-------------
+* API
+  + Add hwloc/windows.h to query Windows processor groups.
+  + Add hwloc_get_obj_with_same_locality() to convert between objects
+    with same locality, for instance NUMA nodes and Packages,
+    or OS devices within a PCI device.
+  + Add hwloc_distances_transform() to modify distances structures.
+    - hwloc-annotate and lstopo have new distances-transform options.
+  + hwloc_distances_add() is replaced with _add_create() followed by
+    _add_values() and _add_commit(). See hwloc/distances.h for details.
+  + Add topology flags to mitigate binding modifications during
+    hwloc discovery, especially on Windows:
+    - HWLOC_TOPOLOGY_FLAG_RESTRICT_TO_CPUBINDING and _MEMBINDING
+      restrict discovery to PUs and NUMA nodes inside the binding.
+    - HWLOC_TOPOLOGY_FLAG_DONT_CHANGE_BINDING prevents from ever
+      changing the binding during discovery.
+* Backends
+  + Add a levelzero backend for oneAPI L0 devices, exposed as OS devices
+    of subtype "LevelZero" and name such as "ze0".
+    - Add hwloc/levelzero.h for interoperability between converting
+      between L0 API devices and hwloc cpusets or OS devices.
+  + Expose NEC Vector Engine cards on Linux as OS devices of subtype
+    "VectorEngine" and name "ve0", etc.
+    Thanks to Anara Kozhokanova, Tim Cramer and Erich Focht for the help.
+  + Add a NVLinkBandwidth distances structure between NVIDIA GPUs
+    (and POWER processor or NVSwitches) in the NVML backend,
+    and a XGMIBandwidth distances structure between AMD GPUs
+    in the RSMI backends.
+    - See "Topology Attributes: Distances, Memory Attributes and CPU Kinds"
+      in the documentation for details about these new distances.
+  + Add support for NUMA node 0 being offline in Linux, thanks to Jirka Hladky.
+* Build
+  + Add --with-cuda-version=<version> or look at the CUDA_VERSION
+    environment variable to find the appropriate CUDA pkg-config files.
+    Thanks to Stephen Herbein for the suggestion.
+    - Also add --with-cuda=<dir> to specify the CUDA installation path
+      manually (and its NVML and OpenCL components).
+      Thanks to Andrea Bocci for the suggestion.
+    - See "How do I enable CUDA and select which CUDA version to use?"
+      in the FAQ for details.
+* Tools
+  + lstopo now has a --windows-processor-groups option on Windows.
+  + hwloc-ps now has a --short-name option to avoid long/truncated
+    command path.
+  + hwloc-ps now has a --single-ancestor option to return a single
+    (possibly too large) object where a process is bound.
+  + hwloc-ps --pid-cmd may now query environment variables,
+    including MPI-specific variables to find out process ranks.
+
+
+Version 2.4.1
+-------------
+* Fix AMD OpenCL device locality when PCI bus or device number >= 128.
+  Thanks to Edgar Leon for reporting the issue.
+  + Applications using any of the following inline functions must
+    be recompiled to get the fix: hwloc_opencl_get_device_pci_busid()
+    hwloc_opencl_get_device_cpuset(), hwloc_opencl_get_device_osdev().
+* Fix the ranking of cpukinds on non-Windows systems,
+  thanks to Ivan Kochin for the report.
+* Fix the insertion of custom Groups after loading the topology,
+  thanks to Scott Hicks.
+* Add support for CPU0 being offline in Linux, thanks to Garrett Clay.
+* Fix missing x86 Package and Core objects FreeBSD/NetBSD.
+  Thanks to Thibault Payet and Yuri Victorovich for the report.
+* Fix the import of very large distances with heterogeneous object types.
+* Fix a memory leak in the Linux backend,
+  thanks to Perceval Anichini.
+
+
 Version 2.4.0
 -------------
 * API

src/3rdparty/hwloc/README

@@ -1,4 +1,8 @@
-Introduction
+This is a truncated and poorly-formatted version of the documentation main page.
+See https://www.open-mpi.org/projects/hwloc/doc/ for more.
+
+
+hwloc Overview
 
 The Hardware Locality (hwloc) software project aims at easing the process of
 discovering hardware resources in parallel architectures. It offers
@@ -8,66 +12,450 @@ high-performance computing (HPC) applications, but is also applicable to any
 project seeking to exploit code and/or data locality on modern computing
 platforms.
 
-hwloc is actually made of two subprojects distributed together:
+hwloc provides command line tools and a C API to obtain the hierarchical map of
+key computing elements within a node, such as: NUMA memory nodes, shared
+caches, processor packages, dies and cores, processing units (logical
+processors or "threads") and even I/O devices. hwloc also gathers various
+attributes such as cache and memory information, and is portable across a
+variety of different operating systems and platforms.
 
-  * The original hwloc project for describing the internals of computing nodes.
+hwloc primarily aims at helping high-performance computing (HPC) applications,
- It is described in details starting at section Hardware Locality (hwloc)
+but is also applicable to any project seeking to exploit code and/or data
- Introduction.
+locality on modern computing platforms.
-  * The network-oriented companion called netloc (Network Locality), described
- in details starting with section Network Locality (netloc).
 
-See also the Related pages tab above for links to other sections.
+hwloc supports the following operating systems:
 
-Netloc may be disabled, but the original hwloc cannot. Both hwloc and netloc
+  * Linux (with knowledge of cgroups and cpusets, memory targets/initiators,
-APIs are documented after these sections.
+ etc.) on all supported hardware, including Intel Xeon Phi, ScaleMP vSMP,
+ and NumaScale NumaConnect.
+  * Solaris (with support for processor sets and logical domains)
+  * AIX
+  * Darwin / OS X
+  * FreeBSD and its variants (such as kFreeBSD/GNU)
+  * NetBSD
+  * HP-UX
+  * Microsoft Windows
+  * IBM BlueGene/Q Compute Node Kernel (CNK)
 
-Installation
+Since it uses standard Operating System information, hwloc's support is mostly
+independant from the processor type (x86, powerpc, ...) and just relies on the
+Operating System support. The main exception is BSD operating systems (NetBSD,
+FreeBSD, etc.) because they do not provide support topology information, hence
+hwloc uses an x86-only CPUID-based backend (which can be used for other OSes
+too, see the Components and plugins section).
 
-hwloc (https://www.open-mpi.org/projects/hwloc/) is available under the BSD
+To check whether hwloc works on a particular machine, just try to build it and
-license. It is hosted as a sub-project of the overall Open MPI project (https:/
+run lstopo or lstopo-no-graphics. If some things do not look right (e.g. bogus
-/www.open-mpi.org/). Note that hwloc does not require any functionality from
+or missing cache information), see Questions and Bugs.
-Open MPI -- it is a wholly separate (and much smaller!) project and code base.
-It just happens to be hosted as part of the overall Open MPI project.
 
-Basic Installation
+hwloc only reports the number of processors on unsupported operating systems;
+no topology information is available.
 
-Installation is the fairly common GNU-based process:
+For development and debugging purposes, hwloc also offers the ability to work
+on "fake" topologies:
 
-shell$ ./configure --prefix=...
+  * Symmetrical tree of resources generated from a list of level arities, see
-shell$ make
+ Synthetic topologies.
-shell$ make install
+  * Remote machine simulation through the gathering of topology as XML files,
+ see Importing and exporting topologies from/to XML files.
 
-hwloc- and netloc-specific configure options and requirements are documented in
+hwloc can display the topology in a human-readable format, either in graphical
-sections hwloc Installation and Netloc Installation respectively.
+mode (X11), or by exporting in one of several different formats, including:
+plain text, LaTeX tikzpicture, PDF, PNG, and FIG (see Command-line Examples
+below). Note that some of the export formats require additional support
+libraries.
 
-Also note that if you install supplemental libraries in non-standard locations,
+hwloc offers a programming interface for manipulating topologies and objects.
-hwloc's configure script may not be able to find them without some help. You
+It also brings a powerful CPU bitmap API that is used to describe topology
-may need to specify additional CPPFLAGS, LDFLAGS, or PKG_CONFIG_PATH values on
+objects location on physical/logical processors. See the Programming Interface
-the configure command line.
+below. It may also be used to binding applications onto certain cores or memory
+nodes. Several utility programs are also provided to ease command-line
+manipulation of topology objects, binding of processes, and so on.
 
-For example, if libpciaccess was installed into /opt/pciaccess, hwloc's
+Bindings for several other languages are available from the project website.
-configure script may not find it be default. Try adding PKG_CONFIG_PATH to the
-./configure command line, like this:
 
-./configure PKG_CONFIG_PATH=/opt/pciaccess/lib/pkgconfig ...
+Command-line Examples
 
-Running the "lstopo" tool is a good way to check as a graphical output whether
+On a 4-package 2-core machine with hyper-threading, the lstopo tool may show
-hwloc properly detected the architecture of your node. Netloc command-line
+the following graphical output:
-tools can be used to display the network topology interconnecting your nodes.
 
-Installing from a Git clone
+[dudley]
 
-Additionally, the code can be directly cloned from Git:
+Here's the equivalent output in textual form:
 
-shell$ git clone https://github.com/open-mpi/hwloc.git
+Machine
-shell$ cd hwloc
+  NUMANode L#0 (P#0)
-shell$ ./autogen.sh
+  Package L#0 + L3 L#0 (4096KB)
+ L2 L#0 (1024KB) + L1 L#0 (16KB) + Core L#0
+   PU L#0 (P#0)
+   PU L#1 (P#8)
+ L2 L#1 (1024KB) + L1 L#1 (16KB) + Core L#1
+   PU L#2 (P#4)
+   PU L#3 (P#12)
+  Package L#1 + L3 L#1 (4096KB)
+ L2 L#2 (1024KB) + L1 L#2 (16KB) + Core L#2
+   PU L#4 (P#1)
+   PU L#5 (P#9)
+ L2 L#3 (1024KB) + L1 L#3 (16KB) + Core L#3
+   PU L#6 (P#5)
+   PU L#7 (P#13)
+  Package L#2 + L3 L#2 (4096KB)
+ L2 L#4 (1024KB) + L1 L#4 (16KB) + Core L#4
+   PU L#8 (P#2)
+   PU L#9 (P#10)
+ L2 L#5 (1024KB) + L1 L#5 (16KB) + Core L#5
+   PU L#10 (P#6)
+   PU L#11 (P#14)
+  Package L#3 + L3 L#3 (4096KB)
+ L2 L#6 (1024KB) + L1 L#6 (16KB) + Core L#6
+   PU L#12 (P#3)
+   PU L#13 (P#11)
+ L2 L#7 (1024KB) + L1 L#7 (16KB) + Core L#7
+   PU L#14 (P#7)
+   PU L#15 (P#15)
 
-Note that GNU Autoconf >=2.63, Automake >=1.11 and Libtool >=2.2.6 are required
+Note that there is also an equivalent output in XML that is meant for exporting
-when building from a Git clone.
+/importing topologies but it is hardly readable to human-beings (see Importing
+and exporting topologies from/to XML files for details).
 
-Nightly development snapshots are available on the web site, they can be
+On a 4-package 2-core Opteron NUMA machine (with two core cores disallowed by
-configured and built without any need for Git or GNU Autotools.
+the administrator), the lstopo tool may show the following graphical output
+(with --disallowed for displaying disallowed objects):
+
+[hagrid]
+
+Here's the equivalent output in textual form:
+
+Machine (32GB total)
+  Package L#0
+ NUMANode L#0 (P#0 8190MB)
+ L2 L#0 (1024KB) + L1 L#0 (64KB) + Core L#0 + PU L#0 (P#0)
+ L2 L#1 (1024KB) + L1 L#1 (64KB) + Core L#1 + PU L#1 (P#1)
+  Package L#1
+ NUMANode L#1 (P#1 8192MB)
+ L2 L#2 (1024KB) + L1 L#2 (64KB) + Core L#2 + PU L#2 (P#2)
+ L2 L#3 (1024KB) + L1 L#3 (64KB) + Core L#3 + PU L#3 (P#3)
+  Package L#2
+ NUMANode L#2 (P#2 8192MB)
+ L2 L#4 (1024KB) + L1 L#4 (64KB) + Core L#4 + PU L#4 (P#4)
+ L2 L#5 (1024KB) + L1 L#5 (64KB) + Core L#5 + PU L#5 (P#5)
+  Package L#3
+ NUMANode L#3 (P#3 8192MB)
+ L2 L#6 (1024KB) + L1 L#6 (64KB) + Core L#6 + PU L#6 (P#6)
+ L2 L#7 (1024KB) + L1 L#7 (64KB) + Core L#7 + PU L#7 (P#7)
+
+On a 2-package quad-core Xeon (pre-Nehalem, with 2 dual-core dies into each
+package):
+
+[emmett]
+
+Here's the same output in textual form:
+
+Machine (total 16GB)
+  NUMANode L#0 (P#0 16GB)
+  Package L#0
+ L2 L#0 (4096KB)
+   L1 L#0 (32KB) + Core L#0 + PU L#0 (P#0)
+   L1 L#1 (32KB) + Core L#1 + PU L#1 (P#4)
+ L2 L#1 (4096KB)
+   L1 L#2 (32KB) + Core L#2 + PU L#2 (P#2)
+   L1 L#3 (32KB) + Core L#3 + PU L#3 (P#6)
+  Package L#1
+ L2 L#2 (4096KB)
+   L1 L#4 (32KB) + Core L#4 + PU L#4 (P#1)
+   L1 L#5 (32KB) + Core L#5 + PU L#5 (P#5)
+ L2 L#3 (4096KB)
+   L1 L#6 (32KB) + Core L#6 + PU L#6 (P#3)
+   L1 L#7 (32KB) + Core L#7 + PU L#7 (P#7)
+
+Programming Interface
+
+The basic interface is available in hwloc.h. Some higher-level functions are
+available in hwloc/helper.h to reduce the need to manually manipulate objects
+and follow links between them. Documentation for all these is provided later in
+this document. Developers may also want to look at hwloc/inlines.h which
+contains the actual inline code of some hwloc.h routines, and at this document,
+which provides good higher-level topology traversal examples.
+
+To precisely define the vocabulary used by hwloc, a Terms and Definitions
+section is available and should probably be read first.
+
+Each hwloc object contains a cpuset describing the list of processing units
+that it contains. These bitmaps may be used for CPU binding and Memory binding.
+hwloc offers an extensive bitmap manipulation interface in hwloc/bitmap.h.
+
+Moreover, hwloc also comes with additional helpers for interoperability with
+several commonly used environments. See the Interoperability With Other
+Software section for details.
+
+The complete API documentation is available in a full set of HTML pages, man
+pages, and self-contained PDF files (formatted for both both US letter and A4
+formats) in the source tarball in doc/doxygen-doc/.
+
+NOTE: If you are building the documentation from a Git clone, you will need to
+have Doxygen and pdflatex installed -- the documentation will be built during
+the normal "make" process. The documentation is installed during "make install"
+to $prefix/share/doc/hwloc/ and your systems default man page tree (under
+$prefix, of course).
+
+Portability
+
+Operating System have varying support for CPU and memory binding, e.g. while
+some Operating Systems provide interfaces for all kinds of CPU and memory
+bindings, some others provide only interfaces for a limited number of kinds of
+CPU and memory binding, and some do not provide any binding interface at all.
+Hwloc's binding functions would then simply return the ENOSYS error (Function
+not implemented), meaning that the underlying Operating System does not provide
+any interface for them. CPU binding and Memory binding provide more information
+on which hwloc binding functions should be preferred because interfaces for
+them are usually available on the supported Operating Systems.
+
+Similarly, the ability of reporting topology information varies from one
+platform to another. As shown in Command-line Examples, hwloc can obtain
+information on a wide variety of hardware topologies. However, some platforms
+and/or operating system versions will only report a subset of this information.
+For example, on an PPC64-based system with 8 cores (each with 2 hardware
+threads) running a default 2.6.18-based kernel from RHEL 5.4, hwloc is only
+able to glean information about NUMA nodes and processor units (PUs). No
+information about caches, packages, or cores is available.
+
+Here's the graphical output from lstopo on this platform when Simultaneous
+Multi-Threading (SMT) is enabled:
+
+[ppc64-with]
+
+And here's the graphical output from lstopo on this platform when SMT is
+disabled:
+
+[ppc64-with]
+
+Notice that hwloc only sees half the PUs when SMT is disabled. PU L#6, for
+example, seems to change location from NUMA node #0 to #1. In reality, no PUs
+"moved" -- they were simply re-numbered when hwloc only saw half as many (see
+also Logical index in Indexes and Sets). Hence, PU L#6 in the SMT-disabled
+picture probably corresponds to PU L#12 in the SMT-enabled picture.
+
+This same "PUs have disappeared" effect can be seen on other platforms -- even
+platforms / OSs that provide much more information than the above PPC64 system.
+This is an unfortunate side-effect of how operating systems report information
+to hwloc.
+
+Note that upgrading the Linux kernel on the same PPC64 system mentioned above
+to 2.6.34, hwloc is able to discover all the topology information. The
+following picture shows the entire topology layout when SMT is enabled:
+
+[ppc64-full]
+
+Developers using the hwloc API or XML output for portable applications should
+therefore be extremely careful to not make any assumptions about the structure
+of data that is returned. For example, per the above reported PPC topology, it
+is not safe to assume that PUs will always be descendants of cores.
+
+Additionally, future hardware may insert new topology elements that are not
+available in this version of hwloc. Long-lived applications that are meant to
+span multiple different hardware platforms should also be careful about making
+structure assumptions. For example, a new element may someday exist between a
+core and a PU.
+
+API Example
+
+The following small C example (available in the source tree as ``doc/examples/
+hwloc-hello.c'') prints the topology of the machine and performs some thread
+and memory binding. More examples are available in the doc/examples/ directory
+of the source tree.
+
+/* Example hwloc API program.
+*
+* See other examples under doc/examples/ in the source tree
+* for more details.
+*
+* Copyright (c) 2009-2016 Inria. All rights reserved.
+* Copyright (c) 2009-2011 Universit?eacute; Bordeaux
+* Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved.
+* See COPYING in top-level directory.
+*
+* hwloc-hello.c
+*/
+#include "hwloc.h"
+#include <errno.h>
+#include <stdio.h>
+#include <string.h>
+static void print_children(hwloc_topology_t topology, hwloc_obj_t obj,
+int depth)
+{
+char type[32], attr[1024];
+unsigned i;
+hwloc_obj_type_snprintf(type, sizeof(type), obj, 0);
+printf("%*s%s", 2*depth, "", type);
+if (obj->os_index != (unsigned) -1)
+printf("#%u", obj->os_index);
+hwloc_obj_attr_snprintf(attr, sizeof(attr), obj, " ", 0);
+if (*attr)
+printf("(%s)", attr);
+printf("\n");
+for (i = 0; i < obj->arity; i++) {
+print_children(topology, obj->children[i], depth + 1);
+}
+}
+int main(void)
+{
+int depth;
+unsigned i, n;
+unsigned long size;
+int levels;
+char string[128];
+int topodepth;
+void *m;
+hwloc_topology_t topology;
+hwloc_cpuset_t cpuset;
+hwloc_obj_t obj;
+/* Allocate and initialize topology object. */
+hwloc_topology_init(&topology);
+/* ... Optionally, put detection configuration here to ignore
+some objects types, define a synthetic topology, etc....
+The default is to detect all the objects of the machine that
+the caller is allowed to access. See Configure Topology
+Detection. */
+/* Perform the topology detection. */
+hwloc_topology_load(topology);
+/* Optionally, get some additional topology information
+in case we need the topology depth later. */
+topodepth = hwloc_topology_get_depth(topology);
+/*****************************************************************
+* First example:
+* Walk the topology with an array style, from level 0 (always
+* the system level) to the lowest level (always the proc level).
+*****************************************************************/
+for (depth = 0; depth < topodepth; depth++) {
+printf("*** Objects at level %d\n", depth);
+for (i = 0; i < hwloc_get_nbobjs_by_depth(topology, depth);
+i++) {
+hwloc_obj_type_snprintf(string, sizeof(string),
+hwloc_get_obj_by_depth(topology, depth, i), 0);
+printf("Index %u: %s\n", i, string);
+}
+}
+/*****************************************************************
+* Second example:
+* Walk the topology with a tree style.
+*****************************************************************/
+printf("*** Printing overall tree\n");
+print_children(topology, hwloc_get_root_obj(topology), 0);
+/*****************************************************************
+* Third example:
+* Print the number of packages.
+*****************************************************************/
+depth = hwloc_get_type_depth(topology, HWLOC_OBJ_PACKAGE);
+if (depth == HWLOC_TYPE_DEPTH_UNKNOWN) {
+printf("*** The number of packages is unknown\n");
+} else {
+printf("*** %u package(s)\n",
+hwloc_get_nbobjs_by_depth(topology, depth));
+}
+/*****************************************************************
+* Fourth example:
+* Compute the amount of cache that the first logical processor
+* has above it.
+*****************************************************************/
+levels = 0;
+size = 0;
+for (obj = hwloc_get_obj_by_type(topology, HWLOC_OBJ_PU, 0);
+obj;
+obj = obj->parent)
+if (hwloc_obj_type_is_cache(obj->type)) {
+levels++;
+size += obj->attr->cache.size;
+}
+printf("*** Logical processor 0 has %d caches totaling %luKB\n",
+levels, size / 1024);
+/*****************************************************************
+* Fifth example:
+* Bind to only one thread of the last core of the machine.
+*
+* First find out where cores are, or else smaller sets of CPUs if
+* the OS doesn't have the notion of a "core".
+*****************************************************************/
+depth = hwloc_get_type_or_below_depth(topology, HWLOC_OBJ_CORE);
+/* Get last core. */
+obj = hwloc_get_obj_by_depth(topology, depth,
+hwloc_get_nbobjs_by_depth(topology, depth) - 1);
+if (obj) {
+/* Get a copy of its cpuset that we may modify. */
+cpuset = hwloc_bitmap_dup(obj->cpuset);
+/* Get only one logical processor (in case the core is
+SMT/hyper-threaded). */
+hwloc_bitmap_singlify(cpuset);
+/* And try to bind ourself there. */
+if (hwloc_set_cpubind(topology, cpuset, 0)) {
+char *str;
+int error = errno;
+hwloc_bitmap_asprintf(&str, obj->cpuset);
+printf("Couldn't bind to cpuset %s: %s\n", str, strerror(error));
+free(str);
+}
+/* Free our cpuset copy */
+hwloc_bitmap_free(cpuset);
+}
+/*****************************************************************
+* Sixth example:
+* Allocate some memory on the last NUMA node, bind some existing
+* memory to the last NUMA node.
+*****************************************************************/
+/* Get last node. There's always at least one. */
+n = hwloc_get_nbobjs_by_type(topology, HWLOC_OBJ_NUMANODE);
+obj = hwloc_get_obj_by_type(topology, HWLOC_OBJ_NUMANODE, n - 1);
+size = 1024*1024;
+m = hwloc_alloc_membind(topology, size, obj->nodeset,
+HWLOC_MEMBIND_BIND, HWLOC_MEMBIND_BYNODESET);
+hwloc_free(topology, m, size);
+m = malloc(size);
+hwloc_set_area_membind(topology, m, size, obj->nodeset,
+HWLOC_MEMBIND_BIND, HWLOC_MEMBIND_BYNODESET);
+free(m);
+/* Destroy topology object. */
+hwloc_topology_destroy(topology);
+return 0;
+}
+
+hwloc provides a pkg-config executable to obtain relevant compiler and linker
+flags. See Compiling software on top of hwloc's C API for details on building
+program on top of hwloc's API using GNU Make or CMake.
+
+On a machine 2 processor packages -- each package of which has two processing
+cores -- the output from running hwloc-hello could be something like the
+following:
+
+shell$ ./hwloc-hello
+*** Objects at level 0
+Index 0: Machine
+*** Objects at level 1
+Index 0: Package#0
+Index 1: Package#1
+*** Objects at level 2
+Index 0: Core#0
+Index 1: Core#1
+Index 2: Core#3
+Index 3: Core#2
+*** Objects at level 3
+Index 0: PU#0
+Index 1: PU#1
+Index 2: PU#2
+Index 3: PU#3
+*** Printing overall tree
+Machine
+  Package#0
+ Core#0
+   PU#0
+ Core#1
+   PU#1
+  Package#1
+ Core#3
+   PU#2
+ Core#2
+   PU#3
+*** 2 package(s)
+*** Logical processor 0 has 0 caches totaling 0KB
+shell$
 
 Questions and Bugs
 
@@ -78,8 +466,22 @@ debug and report issues.
 Questions may be sent to the users or developers mailing lists (https://
 www.open-mpi.org/community/lists/hwloc.php).
 
-There is also a #hwloc IRC channel on Freenode (irc.freenode.net).
+There is also a #hwloc IRC channel on Libera Chat (irc.libera.chat).
+
+History / Credits
+
+hwloc is the evolution and merger of the libtopology project and the Portable
+Linux Processor Affinity (PLPA) (https://www.open-mpi.org/projects/plpa/)
+project. Because of functional and ideological overlap, these two code bases
+and ideas were merged and released under the name "hwloc" as an Open MPI
+sub-project.
+
+libtopology was initially developed by the Inria Runtime Team-Project. PLPA was
+initially developed by the Open MPI development team as a sub-project. Both are
+now deprecated in favor of hwloc, which is distributed as an Open MPI
+sub-project.
 
 
 
-See https://www.open-mpi.org/projects/hwloc/doc/ for more hwloc documentation.
+See https://www.open-mpi.org/projects/hwloc/doc/ for more hwloc documentation,
+actual links to related pages, images, etc.

src/3rdparty/hwloc/VERSION

@@ -8,8 +8,8 @@
 # Please update HWLOC_VERSION* in contrib/windows/hwloc_config.h too.
 
 major=2
-minor=4
+minor=12
-release=0
+release=1
 
 # greek is used for alpha or beta release tags.  If it is non-empty,
 # it will be appended to the version number.  It does not have to be
@@ -22,7 +22,7 @@ greek=
 
 # The date when this release was created
 
-date="Nov 26, 2020"
+date="May 12, 2025"
 
 # If snapshot=1, then use the value from snapshot_version as the
 # entire hwloc version (i.e., ignore major, minor, release, and
@@ -41,7 +41,6 @@ snapshot_version=${major}.${minor}.${release}${greek}-git
 # 2. Version numbers are described in the Libtool current:revision:age
 # format.
 
-libhwloc_so_version=19:0:4
+libhwloc_so_version=25:0:10
-libnetloc_so_version=0:0:0
 
 # Please also update the <TargetName> lines in contrib/windows/libhwloc.vcxproj

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2025 Inria.  All rights reserved.
  * Copyright © 2009-2012 Université Bordeaux
  * Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -11,10 +11,10 @@
 #ifndef HWLOC_CONFIG_H
 #define HWLOC_CONFIG_H
 
-#define HWLOC_VERSION "2.4.0"
+#define HWLOC_VERSION "2.12.1"
 #define HWLOC_VERSION_MAJOR 2
-#define HWLOC_VERSION_MINOR 4
+#define HWLOC_VERSION_MINOR 12
-#define HWLOC_VERSION_RELEASE 0
+#define HWLOC_VERSION_RELEASE 1
 #define HWLOC_VERSION_GREEK ""
 
 #define __hwloc_restrict

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2024 Inria.  All rights reserved.
  * Copyright © 2009-2012 Université Bordeaux
  * Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -50,9 +50,10 @@ extern "C" {
  * hwloc_bitmap_free(set);
  * \endcode
  *
- * \note Most functions below return an int that may be negative in case of
+ * \note Most functions below return 0 on success and -1 on error.
- * error. The usual error case would be an internal failure to realloc/extend
+ * The usual error case would be an internal failure to realloc/extend
  * the storage of the bitmap (\p errno would be set to \c ENOMEM).
+ * See also \ref hwlocality_api_error_reporting.
  *
  * \note Several examples of using the bitmap API are available under the
  * doc/examples/ directory in the source tree.
@@ -83,7 +84,13 @@ typedef const struct hwloc_bitmap_s * hwloc_const_bitmap_t;
  */
 HWLOC_DECLSPEC hwloc_bitmap_t hwloc_bitmap_alloc(void) __hwloc_attribute_malloc;
 
-/** \brief Allocate a new full bitmap. */
+/** \brief Allocate a new full bitmap.
+ *
+ * \returns A valid bitmap or \c NULL.
+ *
+ * The bitmap should be freed by a corresponding call to
+ * hwloc_bitmap_free().
+ */
 HWLOC_DECLSPEC hwloc_bitmap_t hwloc_bitmap_alloc_full(void) __hwloc_attribute_malloc;
 
 /** \brief Free bitmap \p bitmap.
@@ -106,73 +113,134 @@ HWLOC_DECLSPEC int hwloc_bitmap_copy(hwloc_bitmap_t dst, hwloc_const_bitmap_t sr
  * Bitmap/String Conversion
  */
 
-/** \brief Stringify a bitmap.
+/** \brief Stringify a bitmap in the default hwloc format.
+ *
+ * <b>Note that if the bitmap is a CPU or nodeset, it contains physical indexes.</b>
+ *
+ * Print the bits set inside a bitmap as a comma-separated list of hexadecimal 32-bit blocks.
+ * A bitmap containing bits 1, 33, 34, and all from 64 to 95 is printed as <tt>"0xffffffff,0x00000006,0x00000002"</tt>.
  *
  * Up to \p buflen characters may be written in buffer \p buf.
  *
  * If \p buflen is 0, \p buf may safely be \c NULL.
  *
- * \return the number of character that were actually written if not truncating,
+ * \return the number of characters that were actually written if not truncating,
- * or that would have been written (not including the ending \\0).
+ * or that would have been written (not including the ending \c \0).
+ * \return -1 on error.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_snprintf(char * __hwloc_restrict buf, size_t buflen, hwloc_const_bitmap_t bitmap);
 
-/** \brief Stringify a bitmap into a newly allocated string.
+/** \brief Stringify a bitmap into a newly allocated string in the default hwloc format.
  *
- * \return -1 on error.
+ * <b>Note that if the bitmap is a CPU or nodeset, it contains physical indexes.</b>
+ *
+ * Print the bits set inside a bitmap as a comma-separated list of hexadecimal 32-bit blocks.
+ * A bitmap containing bits 1, 33, 34, and all from 64 to 95 is printed as <tt>"0xffffffff,0x00000006,0x00000002"</tt>.
+ *
+ * \return the number of characters that were written (not including the ending \c \0).
+ * \return -1 on error, for instance with \p errno set to \c ENOMEM on failure to allocate the output string.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_asprintf(char ** strp, hwloc_const_bitmap_t bitmap);
 
-/** \brief Parse a bitmap string and stores it in bitmap \p bitmap.
+/** \brief Parse a bitmap string as the default hwloc format and stores it in bitmap \p bitmap.
+ *
+ * <b>Note that if the bitmap is a CPU or nodeset, the input string must contain physical indexes.</b>
+ *
+ * The input string should be a comma-separared list of hexadecimal 32-bit blocks.
+ * String <tt>"0xffffffff,0x6,0x2"</tt> is parsed as a bitmap containing all bits between 64 and 95,
+ * and bits 33, 34 and 1.
+ *
+ * \return 0 on success, -1 on error.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_sscanf(hwloc_bitmap_t bitmap, const char * __hwloc_restrict string);
 
 /** \brief Stringify a bitmap in the list format.
+ *
+ * <b>Note that if the bitmap is a CPU or nodeset, it contains physical indexes.</b>
  *
  * Lists are comma-separated indexes or ranges.
  * Ranges are dash separated indexes.
- * The last range may not have an ending indexes if the bitmap is infinitely set.
+ * A bitmap containing bits 1, 33, 34, and all from 64 to 95 is printed as <tt>"1,33-34,64-95"</tt>.
+ * The last range may not have an ending index if the bitmap is infinitely set.
  *
  * Up to \p buflen characters may be written in buffer \p buf.
  *
  * If \p buflen is 0, \p buf may safely be \c NULL.
  *
- * \return the number of character that were actually written if not truncating,
+ * \return the number of characters that were actually written if not truncating,
- * or that would have been written (not including the ending \\0).
+ * or that would have been written (not including the ending \c \0).
+ * \return -1 on error.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_list_snprintf(char * __hwloc_restrict buf, size_t buflen, hwloc_const_bitmap_t bitmap);
 
 /** \brief Stringify a bitmap into a newly allocated list string.
  *
- * \return -1 on error.
+ * <b>Note that if the bitmap is a CPU or nodeset, it contains physical indexes.</b>
+ *
+ * Lists are comma-separated indexes or ranges.
+ * Ranges are dash separated indexes.
+ * A bitmap containing bits 1, 33, 34, and all from 64 to 95 is printed as <tt>"1,33-34,64-95"</tt>.
+ * The last range may not have an ending index if the bitmap is infinitely set.
+ *
+ * \return the number of characters that were written (not including the ending \c \0).
+ * \return -1 on error, for instance with \p errno set to \c ENOMEM on failure to allocate the output string.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_list_asprintf(char ** strp, hwloc_const_bitmap_t bitmap);
 
 /** \brief Parse a list string and stores it in bitmap \p bitmap.
+ *
+ * <b>Note that if the bitmap is a CPU or nodeset, the input string must contain physical indexes.</b>
+ *
+ * Lists are comma-separated indexes or ranges.
+ * Ranges are dash separated indexes.
+ * String <tt>"1,33-34,64-95"</tt> is parsed as a bitmap containing bits 1, 33, 34, and all from 64 to 95.
+ * The last range may not have an ending index if the bitmap is infinitely set.
+ *
+ * \return 0 on success, -1 on error.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_list_sscanf(hwloc_bitmap_t bitmap, const char * __hwloc_restrict string);
 
 /** \brief Stringify a bitmap in the taskset-specific format.
  *
- * The taskset command manipulates bitmap strings that contain a single
+ * <b>Note that if the bitmap is a CPU or nodeset, it contains physical indexes.</b>
+ *
+ * The taskset program manipulates bitmap strings that contain a single
  * (possible very long) hexadecimal number starting with 0x.
+ * A bitmap containing bits 1, 33, 34, and all from 64 to 95 is printed as </tt>"0xffffffff0000000600000002"</tt>.
  *
  * Up to \p buflen characters may be written in buffer \p buf.
  *
  * If \p buflen is 0, \p buf may safely be \c NULL.
  *
- * \return the number of character that were actually written if not truncating,
+ * \return the number of characters that were actually written if not truncating,
- * or that would have been written (not including the ending \\0).
+ * or that would have been written (not including the ending \c \0).
+ * \return -1 on error.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_taskset_snprintf(char * __hwloc_restrict buf, size_t buflen, hwloc_const_bitmap_t bitmap);
 
 /** \brief Stringify a bitmap into a newly allocated taskset-specific string.
  *
- * \return -1 on error.
+ * <b>Note that if the bitmap is a CPU or nodeset, it contains physical indexes.</b>
+ *
+ * The taskset program manipulates bitmap strings that contain a single
+ * (possible very long) hexadecimal number starting with 0x.
+ * A bitmap containing bits 1, 33, 34, and all from 64 to 95 is printed as <tt>"0xffffffff0000000600000002"</tt>.
+ *
+ * \return the number of characters that were written (not including the ending \c \0).
+ * \return -1 on error, for instance with \p errno set to \c ENOMEM on failure to allocate the output string.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_taskset_asprintf(char ** strp, hwloc_const_bitmap_t bitmap);
 
 /** \brief Parse a taskset-specific bitmap string and stores it in bitmap \p bitmap.
+ *
+ * <b>Note that if the bitmap is a CPU or nodeset, the input string must contain physical indexes.</b>
+ *
+ * The taskset program manipulates bitmap strings that contain a single
+ * (possible very long) hexadecimal number starting with 0x.
+ * String <tt>"0xffffffff0000000600000002"</tt> is parsed as a bitmap containing all bits between 64 and 95,
+ * and bits 33, 34 and 1.
+ *
+ * \return 0 on success, -1 on error.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_taskset_sscanf(hwloc_bitmap_t bitmap, const char * __hwloc_restrict string);
 
@@ -279,6 +347,7 @@ HWLOC_DECLSPEC int hwloc_bitmap_to_ulongs(hwloc_const_bitmap_t bitmap, unsigned
  * When called on the output of hwloc_topology_get_topology_cpuset(),
  * the returned number is large enough for all cpusets of the topology.
  *
+ * \return the number of unsigned longs required.
  * \return -1 if \p bitmap is infinite.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_nr_ulongs(hwloc_const_bitmap_t bitmap) __hwloc_attribute_pure;
@@ -305,21 +374,23 @@ HWLOC_DECLSPEC int hwloc_bitmap_isfull(hwloc_const_bitmap_t bitmap) __hwloc_attr
 
 /** \brief Compute the first index (least significant bit) in bitmap \p bitmap
  *
- * \return -1 if no index is set in \p bitmap.
+ * \return the first index set in \p bitmap.
+ * \return -1 if \p bitmap is empty.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_first(hwloc_const_bitmap_t bitmap) __hwloc_attribute_pure;
 
 /** \brief Compute the next index in bitmap \p bitmap which is after index \p prev
  *
- * If \p prev is -1, the first index is returned.
+ * \return the first index set in \p bitmap if \p prev is \c -1.
- *
+ * \return the next index set in \p bitmap if \p prev is not \c -1.
  * \return -1 if no index with higher index is set in \p bitmap.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_next(hwloc_const_bitmap_t bitmap, int prev) __hwloc_attribute_pure;
 
 /** \brief Compute the last index (most significant bit) in bitmap \p bitmap
  *
- * \return -1 if no index is set in \p bitmap, or if \p bitmap is infinitely set.
+ * \return the last index set in \p bitmap.
+ * \return -1 if \p bitmap is empty, or if \p bitmap is infinitely set.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_last(hwloc_const_bitmap_t bitmap) __hwloc_attribute_pure;
 
@@ -327,28 +398,29 @@ HWLOC_DECLSPEC int hwloc_bitmap_last(hwloc_const_bitmap_t bitmap) __hwloc_attrib
  * indexes that are in the bitmap).
  *
  * \return the number of indexes that are in the bitmap.
- *
  * \return -1 if \p bitmap is infinitely set.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_weight(hwloc_const_bitmap_t bitmap) __hwloc_attribute_pure;
 
 /** \brief Compute the first unset index (least significant bit) in bitmap \p bitmap
  *
- * \return -1 if no index is unset in \p bitmap.
+ * \return the first unset index in \p bitmap.
+ * \return -1 if \p bitmap is full.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_first_unset(hwloc_const_bitmap_t bitmap) __hwloc_attribute_pure;
 
 /** \brief Compute the next unset index in bitmap \p bitmap which is after index \p prev
  *
- * If \p prev is -1, the first unset index is returned.
+ * \return the first index unset in \p bitmap if \p prev is \c -1.
- *
+ * \return the next index unset in \p bitmap if \p prev is not \c -1.
  * \return -1 if no index with higher index is unset in \p bitmap.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_next_unset(hwloc_const_bitmap_t bitmap, int prev) __hwloc_attribute_pure;
 
 /** \brief Compute the last unset index (most significant bit) in bitmap \p bitmap
  *
- * \return -1 if no index is unset in \p bitmap, or if \p bitmap is infinitely set.
+ * \return the last index unset in \p bitmap.
+ * \return -1 if \p bitmap is full, or if \p bitmap is not infinitely set.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_last_unset(hwloc_const_bitmap_t bitmap) __hwloc_attribute_pure;
 
@@ -357,11 +429,11 @@ HWLOC_DECLSPEC int hwloc_bitmap_last_unset(hwloc_const_bitmap_t bitmap) __hwloc_
  * The loop must start with hwloc_bitmap_foreach_begin() and end
  * with hwloc_bitmap_foreach_end() followed by a terminating ';'.
  *
- * \p index is the loop variable; it should be an unsigned int.  The
+ * \p id is the loop variable; it should be an unsigned int.  The
- * first iteration will set \p index to the lowest index in the bitmap.
+ * first iteration will set \p id to the lowest index in the bitmap.
  * Successive iterations will iterate through, in order, all remaining
  * indexes set in the bitmap.  To be specific: each iteration will return a
- * value for \p index such that hwloc_bitmap_isset(bitmap, index) is true.
+ * value for \p id such that hwloc_bitmap_isset(bitmap, id) is true.
  *
  * The assert prevents the loop from being infinite if the bitmap is infinitely set.
  *
@@ -428,6 +500,8 @@ HWLOC_DECLSPEC int hwloc_bitmap_not (hwloc_bitmap_t res, hwloc_const_bitmap_t bi
 /** \brief Test whether bitmaps \p bitmap1 and \p bitmap2 intersects.
  *
  * \return 1 if bitmaps intersect, 0 otherwise.
+ *
+ * \note The empty bitmap does not intersect any other bitmap.
  */
 HWLOC_DECLSPEC int hwloc_bitmap_intersects (hwloc_const_bitmap_t bitmap1, hwloc_const_bitmap_t bitmap2) __hwloc_attribute_pure;
 

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

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

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

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2010-2020 Inria.  All rights reserved.
+ * Copyright © 2010-2023 Inria.  All rights reserved.
  * Copyright © 2010-2011 Université Bordeaux
  * Copyright © 2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -42,6 +42,9 @@ extern "C" {
 /** \brief Return the domain, bus and device IDs of the CUDA device \p cudevice.
  *
  * Device \p cudevice must match the local machine.
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if device information could not be found.
  */
 static __hwloc_inline int
 hwloc_cuda_get_device_pci_ids(hwloc_topology_t topology __hwloc_attribute_unused,
@@ -75,7 +78,7 @@ hwloc_cuda_get_device_pci_ids(hwloc_topology_t topology __hwloc_attribute_unused
 /** \brief Get the CPU set of processors that are physically
  * close to device \p cudevice.
  *
- * Return the CPU set describing the locality of the CUDA device \p cudevice.
+ * Store in \p set the CPU-set describing the locality of the CUDA device \p cudevice.
  *
  * Topology \p topology and device \p cudevice must match the local machine.
  * I/O devices detection and the CUDA component are not needed in the topology.
@@ -87,6 +90,9 @@ hwloc_cuda_get_device_pci_ids(hwloc_topology_t topology __hwloc_attribute_unused
  *
  * This function is currently only implemented in a meaningful way for
  * Linux; other systems will simply get a full cpuset.
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if device information could not be found.
  */
 static __hwloc_inline int
 hwloc_cuda_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
@@ -120,8 +126,8 @@ hwloc_cuda_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
 /** \brief Get the hwloc PCI device object corresponding to the
  * CUDA device \p cudevice.
  *
- * Return the PCI device object describing the CUDA device \p cudevice.
+ * \return The hwloc PCI device object describing the CUDA device \p cudevice.
- * Return NULL if there is none.
+ * \return \c NULL if none could be found.
  *
  * Topology \p topology and device \p cudevice must match the local machine.
  * I/O devices detection must be enabled in topology \p topology.
@@ -140,8 +146,8 @@ hwloc_cuda_get_device_pcidev(hwloc_topology_t topology, CUdevice cudevice)
 
 /** \brief Get the hwloc OS device object corresponding to CUDA device \p cudevice.
  *
- * Return the hwloc OS device object that describes the given
+ * \return The hwloc OS device object that describes the given CUDA device \p cudevice.
- * CUDA device \p cudevice. Return NULL if there is none.
+ * \return \c NULL if none could be found.
  *
  * Topology \p topology and device \p cudevice must match the local machine.
  * I/O devices detection and the CUDA component must be enabled in the topology.
@@ -183,8 +189,8 @@ hwloc_cuda_get_device_osdev(hwloc_topology_t topology, CUdevice cudevice)
 /** \brief Get the hwloc OS device object corresponding to the
  * CUDA device whose index is \p idx.
  *
- * Return the OS device object describing the CUDA device whose
+ * \return The hwloc OS device object describing the CUDA device whose index is \p idx.
- * index is \p idx. Return NULL if there is none.
+ * \return \c NULL if none could be found.
  *
  * The topology \p topology does not necessarily have to match the current
  * machine. For instance the topology may be an XML import of a remote host.

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

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2010-2020 Inria.  All rights reserved.
+ * Copyright © 2010-2023 Inria.  All rights reserved.
  * Copyright © 2010-2011 Université Bordeaux
  * Copyright © 2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -43,6 +43,9 @@ extern "C" {
 /** \brief Return the domain, bus and device IDs of the CUDA device whose index is \p idx.
  *
  * Device index \p idx must match the local machine.
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if device information could not be found.
  */
 static __hwloc_inline int
 hwloc_cudart_get_device_pci_ids(hwloc_topology_t topology __hwloc_attribute_unused,
@@ -72,7 +75,7 @@ hwloc_cudart_get_device_pci_ids(hwloc_topology_t topology __hwloc_attribute_unus
 /** \brief Get the CPU set of processors that are physically
  * close to device \p idx.
  *
- * Return the CPU set describing the locality of the CUDA device
+ * Store in \p set the CPU-set describing the locality of the CUDA device
  * whose index is \p idx.
  *
  * Topology \p topology and device \p idx must match the local machine.
@@ -84,6 +87,9 @@ hwloc_cudart_get_device_pci_ids(hwloc_topology_t topology __hwloc_attribute_unus
  *
  * This function is currently only implemented in a meaningful way for
  * Linux; other systems will simply get a full cpuset.
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if device information could not be found.
  */
 static __hwloc_inline int
 hwloc_cudart_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
@@ -117,8 +123,8 @@ hwloc_cudart_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unuse
 /** \brief Get the hwloc PCI device object corresponding to the
  * CUDA device whose index is \p idx.
  *
- * Return the PCI device object describing the CUDA device whose
+ * \return The hwloc PCI device object describing the CUDA device whose index is \p idx.
- * index is \p idx. Return NULL if there is none.
+ * \return \c NULL if none could be found.
  *
  * Topology \p topology and device \p idx must match the local machine.
  * I/O devices detection must be enabled in topology \p topology.
@@ -138,8 +144,8 @@ hwloc_cudart_get_device_pcidev(hwloc_topology_t topology, int idx)
 /** \brief Get the hwloc OS device object corresponding to the
  * CUDA device whose index is \p idx.
  *
- * Return the OS device object describing the CUDA device whose
+ * \return The hwloc OS device object describing the CUDA device whose index is \p idx.
- * index is \p idx. Return NULL if there is none.
+ * \return \c NULL if none could be found.
  *
  * The topology \p topology does not necessarily have to match the current
  * machine. For instance the topology may be an XML import of a remote host.

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2018 Inria.  All rights reserved.
+ * Copyright © 2009-2022 Inria.  All rights reserved.
  * Copyright © 2009-2012 Université Bordeaux
  * Copyright © 2009-2010 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -30,6 +30,15 @@ extern "C" {
 /* backward compat with v1.10 before Node->NUMANode clarification */
 #define HWLOC_OBJ_NODE HWLOC_OBJ_NUMANODE
 
+/** \brief Add a distances structure.
+ *
+ * Superseded by hwloc_distances_add_create()+hwloc_distances_add_values()+hwloc_distances_add_commit()
+ * in v2.5.
+ */
+HWLOC_DECLSPEC int hwloc_distances_add(hwloc_topology_t topology,
+				       unsigned nbobjs, hwloc_obj_t *objs, hwloc_uint64_t *values,
+				       unsigned long kind, unsigned long flags) __hwloc_attribute_deprecated;
+
 /** \brief Insert a misc object by parent.
  *
  * Identical to hwloc_topology_insert_misc_object().
@@ -46,7 +55,7 @@ hwloc_topology_insert_misc_object_by_parent(hwloc_topology_t topology, hwloc_obj
  *
  * If \p size is 0, \p string may safely be \c NULL.
  *
- * \return the number of character that were actually written if not truncating,
+ * \return the number of characters that were actually written if not truncating,
  * or that would have been written (not including the ending \\0).
  */
 static __hwloc_inline int

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

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2013-2020 Inria.  All rights reserved.
+ * Copyright © 2013-2024 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -222,6 +222,8 @@ enum hwloc_topology_diff_apply_flags_e {
 HWLOC_DECLSPEC int hwloc_topology_diff_apply(hwloc_topology_t topology, hwloc_topology_diff_t diff, unsigned long flags);
 
 /** \brief Destroy a list of topology differences.
+ *
+ * \return 0.
  */
 HWLOC_DECLSPEC int hwloc_topology_diff_destroy(hwloc_topology_diff_t diff);
 
@@ -233,6 +235,8 @@ HWLOC_DECLSPEC int hwloc_topology_diff_destroy(hwloc_topology_diff_t diff);
  * This identifier is usually the name of the other XML file
  * that contains the reference topology.
  *
+ * \return 0 on success, -1 on error.
+ *
  * \note the pointer returned in refname should later be freed
  * by the caller.
  */
@@ -246,10 +250,17 @@ HWLOC_DECLSPEC int hwloc_topology_diff_load_xml(const char *xmlpath, hwloc_topol
  * This identifier is usually the name of the other XML file
  * that contains the reference topology.
  * This attribute is given back when reading the diff from XML.
+ *
+ * \return 0 on success, -1 on error.
  */
 HWLOC_DECLSPEC int hwloc_topology_diff_export_xml(hwloc_topology_diff_t diff, const char *refname, const char *xmlpath);
 
 /** \brief Load a list of topology differences from a XML buffer.
+ *
+ * Build a list of differences from the XML memory buffer given
+ * at \p xmlbuffer and of length \p buflen (including an ending \c \0).
+ * This buffer may have been filled earlier with
+ * hwloc_topology_diff_export_xmlbuffer().
  *
  * If not \c NULL, \p refname will be filled with the identifier
  * string of the reference topology for the difference file,
@@ -257,6 +268,8 @@ HWLOC_DECLSPEC int hwloc_topology_diff_export_xml(hwloc_topology_diff_t diff, co
  * This identifier is usually the name of the other XML file
  * that contains the reference topology.
  *
+ * \return 0 on success, -1 on error.
+ *
  * \note the pointer returned in refname should later be freed
  * by the caller.
   */
@@ -271,9 +284,11 @@ HWLOC_DECLSPEC int hwloc_topology_diff_load_xmlbuffer(const char *xmlbuffer, int
  * that contains the reference topology.
  * This attribute is given back when reading the diff from XML.
  *
- * The returned buffer ends with a \0 that is included in the returned
+ * The returned buffer ends with a \c \0 that is included in the returned
  * length.
  *
+ * \return 0 on success, -1 on error.
+ *
  * \note The XML buffer should later be freed with hwloc_free_xmlbuffer().
  */
 HWLOC_DECLSPEC int hwloc_topology_diff_export_xmlbuffer(hwloc_topology_diff_t diff, const char *refname, char **xmlbuffer, int *buflen);

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

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2010-2020 Inria.  All rights reserved.
+ * Copyright © 2010-2025 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -28,16 +28,27 @@ extern "C" {
 
 /** \brief Matrix of distances between a set of objects.
  *
- * This matrix often contains latencies between NUMA nodes
+ * The most common matrix contains latencies between NUMA nodes
  * (as reported in the System Locality Distance Information Table (SLIT)
  * in the ACPI specification), which may or may not be physically accurate.
  * It corresponds to the latency for accessing the memory of one node
  * from a core in another node.
- * The corresponding kind is ::HWLOC_DISTANCES_KIND_FROM_OS | ::HWLOC_DISTANCES_KIND_FROM_USER.
+ * The corresponding kind is ::HWLOC_DISTANCES_KIND_MEANS_LATENCY | ::HWLOC_DISTANCES_KIND_FROM_USER.
  * The name of this distances structure is "NUMALatency".
  *
  * The matrix may also contain bandwidths between random sets of objects,
  * possibly provided by the user, as specified in the \p kind attribute.
+ * Others common distance structures include and "XGMIBandwidth", "XGMIHops",
+ * "XeLinkBandwidth" and "NVLinkBandwidth".
+ *
+ * Pointers \p objs and \p values should not be replaced, reallocated, freed, etc.
+ * However callers are allowed to modify \p kind as well as the contents
+ * of \p objs and \p values arrays.
+ * For instance, if there is a single NUMA node per Package,
+ * hwloc_get_obj_with_same_locality() may be used to convert between them
+ * and replace NUMA nodes in the \p objs array with the corresponding Packages.
+ * See also hwloc_distances_transform() for applying some transformations
+ * to the structure.
  */
 struct hwloc_distances_s {
   unsigned nbobjs;		/**< \brief Number of objects described by the distance matrix. */
@@ -59,11 +70,10 @@ struct hwloc_distances_s {
  * The \p kind attribute of struct hwloc_distances_s is a OR'ed set
  * of kinds.
  *
- * A kind of format HWLOC_DISTANCES_KIND_FROM_* specifies where the
+ * Each distance matrix may have only one kind among HWLOC_DISTANCES_KIND_FROM_*
- * distance information comes from, if known.
+ * specifying where distance information comes from,
- *
+ * and one kind among HWLOC_DISTANCES_KIND_MEANS_* specifying
- * A kind of format HWLOC_DISTANCES_KIND_MEANS_* specifies whether
+ * whether values are latencies or bandwidths.
- * values are latencies or bandwidths, if applicable.
  */
 enum hwloc_distances_kind_e {
   /** \brief These distances were obtained from the operating system or hardware.
@@ -91,6 +101,8 @@ enum hwloc_distances_kind_e {
   HWLOC_DISTANCES_KIND_MEANS_BANDWIDTH = (1UL<<3),
 
   /** \brief This distances structure covers objects of different types.
+   * This may apply to the "NVLinkBandwidth" structure in presence
+   * of a NVSwitch or POWER processor NVLink port.
    * \hideinitializer
    */
   HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES = (1UL<<4)
@@ -118,6 +130,8 @@ enum hwloc_distances_kind_e {
  *
  * Each distance matrix returned in the \p distances array should be released
  * by the caller using hwloc_distances_release().
+ *
+ * \return 0 on success, -1 on error.
  */
 HWLOC_DECLSPEC int
 hwloc_distances_get(hwloc_topology_t topology,
@@ -127,6 +141,8 @@ hwloc_distances_get(hwloc_topology_t topology,
 /** \brief Retrieve distance matrices for object at a specific depth in the topology.
  *
  * Identical to hwloc_distances_get() with the additional \p depth filter.
+ *
+ * \return 0 on success, -1 on error.
  */
 HWLOC_DECLSPEC int
 hwloc_distances_get_by_depth(hwloc_topology_t topology, int depth,
@@ -136,6 +152,8 @@ hwloc_distances_get_by_depth(hwloc_topology_t topology, int depth,
 /** \brief Retrieve distance matrices for object of a specific type.
  *
  * Identical to hwloc_distances_get() with the additional \p type filter.
+ *
+ * \return 0 on success, -1 on error.
  */
 HWLOC_DECLSPEC int
 hwloc_distances_get_by_type(hwloc_topology_t topology, hwloc_obj_type_t type,
@@ -147,6 +165,10 @@ hwloc_distances_get_by_type(hwloc_topology_t topology, hwloc_obj_type_t type,
  * Usually only one distances structure may match a given name.
  *
  * The name of the most common structure is "NUMALatency".
+ * Others include "XGMIBandwidth", "XGMIHops", "XeLinkBandwidth",
+ * and "NVLinkBandwidth".
+ *
+ * \return 0 on success, -1 on error.
  */
 HWLOC_DECLSPEC int
 hwloc_distances_get_by_name(hwloc_topology_t topology, const char *name,
@@ -156,7 +178,12 @@ hwloc_distances_get_by_name(hwloc_topology_t topology, const char *name,
 /** \brief Get a description of what a distances structure contains.
  *
  * For instance "NUMALatency" for hardware-provided NUMA distances (ACPI SLIT),
- * or NULL if unknown.
+ * or \c NULL if unknown.
+ *
+ * \return the constant string with the name of the distance structure.
+ *
+ * \note The returned name should not be freed by the caller,
+ * it belongs to the hwloc library.
  */
 HWLOC_DECLSPEC const char *
 hwloc_distances_get_name(hwloc_topology_t topology, struct hwloc_distances_s *distances);
@@ -168,6 +195,94 @@ hwloc_distances_get_name(hwloc_topology_t topology, struct hwloc_distances_s *di
 HWLOC_DECLSPEC void
 hwloc_distances_release(hwloc_topology_t topology, struct hwloc_distances_s *distances);
 
+/** \brief Transformations of distances structures. */
+enum hwloc_distances_transform_e {
+  /** \brief Remove \c NULL objects from the distances structure.
+   *
+   * Every object that was replaced with \c NULL in the \p objs array
+   * is removed and the \p values array is updated accordingly.
+   *
+   * At least \c 2 objects must remain, otherwise hwloc_distances_transform()
+   * will return \c -1 with \p errno set to \c EINVAL.
+   *
+   * \p kind will be updated with or without ::HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES
+   * according to the remaining objects.
+   *
+   * \hideinitializer
+   */
+  HWLOC_DISTANCES_TRANSFORM_REMOVE_NULL = 0,
+
+  /** \brief Replace bandwidth values with a number of links.
+   *
+   * Usually all values will be either \c 0 (no link) or \c 1 (one link).
+   * However some matrices could get larger values if some pairs of
+   * peers are connected by different numbers of links.
+   *
+   * Values on the diagonal are set to \c 0.
+   *
+   * This transformation only applies to bandwidth matrices.
+   *
+   * \hideinitializer
+   */
+  HWLOC_DISTANCES_TRANSFORM_LINKS = 1,
+
+  /** \brief Merge switches with multiple ports into a single object.
+   *
+   * This currently only applies to NVSwitches where GPUs seem connected
+   * to different switch ports. Switch ports must be objects with subtype
+   * "NVSwitch" as in the NVLinkBandwidth matrix.
+   *
+   * This transformation will replace all ports with only the first one,
+   * now connected to all GPUs. Other ports are removed by applying
+   * ::HWLOC_DISTANCES_TRANSFORM_REMOVE_NULL internally.
+   * \hideinitializer
+   */
+  HWLOC_DISTANCES_TRANSFORM_MERGE_SWITCH_PORTS = 2,
+
+  /** \brief Apply a transitive closure to the matrix to connect objects across switches.
+   *
+   * All pairs of GPUs will be reported as directly connected instead GPUs being
+   * only connected to switches.
+   *
+   * Switch ports must be objects with subtype "NVSwitch" as in the NVLinkBandwidth matrix.
+   * \hideinitializer
+   */
+  HWLOC_DISTANCES_TRANSFORM_TRANSITIVE_CLOSURE = 3
+};
+
+/** \brief Apply a transformation to a distances structure.
+ *
+ * Modify a distances structure that was previously obtained with
+ * hwloc_distances_get() or one of its variants.
+ *
+ * This modifies the local copy of the distances structures but does
+ * not modify the distances information stored inside the topology
+ * (retrieved by another call to hwloc_distances_get() or exported to XML).
+ * To do so, one should add a new distances structure with same
+ * name, kind, objects and values (see \ref hwlocality_distances_add)
+ * and then remove this old one with hwloc_distances_release_remove().
+ *
+ * \p transform must be one of the transformations listed
+ * in ::hwloc_distances_transform_e.
+ *
+ * These transformations may modify the contents of the \p objs or \p values arrays.
+ *
+ * \p transform_attr must be \c NULL for now.
+ *
+ * \p flags must be \c 0 for now.
+ *
+ * \return 0 on success, -1 on error for instance if flags are invalid.
+ *
+ * \note Objects in distances array \p objs may be directly modified
+ * in place without using hwloc_distances_transform().
+ * One may use hwloc_get_obj_with_same_locality() to easily convert
+ * between similar objects of different types.
+ */
+HWLOC_DECLSPEC int hwloc_distances_transform(hwloc_topology_t topology, struct hwloc_distances_s *distances,
+                                             enum hwloc_distances_transform_e transform,
+                                             void *transform_attr,
+                                             unsigned long flags);
+
 /** @} */
 
 
@@ -178,6 +293,7 @@ hwloc_distances_release(hwloc_topology_t topology, struct hwloc_distances_s *dis
 
 /** \brief Find the index of an object in a distances structure.
  *
+ * \return the index of the object in the distances structure if any.
  * \return -1 if object \p obj is not involved in structure \p distances.
  */
 static __hwloc_inline int
@@ -195,6 +311,7 @@ hwloc_distances_obj_index(struct hwloc_distances_s *distances, hwloc_obj_t obj)
  * The distance from \p obj1 to \p obj2 is stored in the value pointed by
  * \p value1to2 and reciprocally.
  *
+ * \return 0 on success.
  * \return -1 if object \p obj1 or \p obj2 is not involved in structure \p distances.
  */
 static __hwloc_inline int
@@ -215,13 +332,87 @@ hwloc_distances_obj_pair_values(struct hwloc_distances_s *distances,
 
 
 
-/** \defgroup hwlocality_distances_add Add or remove distances between objects
+/** \defgroup hwlocality_distances_add Add distances between objects
+ *
+ * The usual way to add distances is:
+ * \code
+ * hwloc_distances_add_handle_t handle;
+ * int err = -1;
+ * handle = hwloc_distances_add_create(topology, "name", kind, 0);
+ * if (handle) {
+ *   err = hwloc_distances_add_values(topology, handle, nbobjs, objs, values, 0);
+ *   if (!err)
+ *     err = hwloc_distances_add_commit(topology, handle, flags);
+ * }
+ * \endcode
+ * If \p err is \c 0 at the end, then addition was successful.
+ *
  * @{
  */
 
+/** \brief Handle to a new distances structure during its addition to the topology. */
+typedef void * hwloc_distances_add_handle_t;
+
+/** \brief Create a new empty distances structure.
+ *
+ * Create an empty distances structure
+ * to be filled with hwloc_distances_add_values()
+ * and then committed with hwloc_distances_add_commit().
+ *
+ * Parameter \p name is optional, it may be \c NULL.
+ * Otherwise, it will be copied internally and may later be freed by the caller.
+ *
+ * \p kind specifies the kind of distance as a OR'ed set of ::hwloc_distances_kind_e.
+ * Only one kind of meaning and one kind of provenance may be given if appropriate
+ * (e.g. ::HWLOC_DISTANCES_KIND_MEANS_BANDWIDTH and ::HWLOC_DISTANCES_KIND_FROM_USER).
+ * Kind ::HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES will be automatically set
+ * according to objects having different types in hwloc_distances_add_values().
+ *
+ * \p flags must be \c 0 for now.
+ *
+ * \return A hwloc_distances_add_handle_t that should then be passed
+ * to hwloc_distances_add_values() and hwloc_distances_add_commit().
+ *
+ * \return \c NULL on error.
+ */
+HWLOC_DECLSPEC hwloc_distances_add_handle_t
+hwloc_distances_add_create(hwloc_topology_t topology,
+                           const char *name, unsigned long kind,
+                           unsigned long flags);
+
+/** \brief Specify the objects and values in a new empty distances structure.
+ *
+ * Specify the objects and values for a new distances structure
+ * that was returned as a handle by hwloc_distances_add_create().
+ * The structure must then be committed with hwloc_distances_add_commit().
+ *
+ * The number of objects is \p nbobjs and the array of objects is \p objs.
+ * Distance values are stored as a one-dimension array in \p values.
+ * The distance from object i to object j is in slot i*nbobjs+j.
+ *
+ * \p nbobjs must be at least 2.
+ *
+ * Arrays \p objs and \p values will be copied internally,
+ * they may later be freed by the caller.
+ *
+ * On error, the temporary distances structure and its content are destroyed.
+ *
+ * \p flags must be \c 0 for now.
+ *
+ * \return 0 on success.
+ * \return -1 on error.
+ */
+HWLOC_DECLSPEC int hwloc_distances_add_values(hwloc_topology_t topology,
+                                              hwloc_distances_add_handle_t handle,
+                                              unsigned nbobjs, hwloc_obj_t *objs,
+                                              hwloc_uint64_t *values,
+                                              unsigned long flags);
+
 /** \brief Flags for adding a new distances to a topology. */
 enum hwloc_distances_add_flag_e {
   /** \brief Try to group objects based on the newly provided distance information.
+   * Grouping is only performed when the distances structure contains latencies,
+   * and when all objects are of the same type.
    * \hideinitializer
    */
   HWLOC_DISTANCES_ADD_FLAG_GROUP = (1UL<<0),
@@ -233,23 +424,33 @@ enum hwloc_distances_add_flag_e {
   HWLOC_DISTANCES_ADD_FLAG_GROUP_INACCURATE = (1UL<<1)
 };
 
-/** \brief Provide a new distance matrix.
+/** \brief Commit a new distances structure.
  *
- * Provide the matrix of distances between a set of objects given by \p nbobjs
+ * This function finalizes the distances structure and inserts in it the topology.
- * and the \p objs array. \p nbobjs must be at least 2.
- * The distances are stored as a one-dimension array in \p values.
- * The distance from object i to object j is in slot i*nbobjs+j.
  *
- * \p kind specifies the kind of distance as a OR'ed set of ::hwloc_distances_kind_e.
+ * Parameter \p handle was previously returned by hwloc_distances_add_create().
- * Kind ::HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES will be automatically added
+ * Then objects and values were specified with hwloc_distances_add_values().
- * if objects of different types are given.
  *
  * \p flags configures the behavior of the function using an optional OR'ed set of
  * ::hwloc_distances_add_flag_e.
+ * It may be used to request the grouping of existing objects based on distances.
+ *
+ * On error, the temporary distances structure and its content are destroyed.
+ *
+ * \return 0 on success.
+ * \return -1 on error.
+ */
+HWLOC_DECLSPEC int hwloc_distances_add_commit(hwloc_topology_t topology,
+                                              hwloc_distances_add_handle_t handle,
+                                              unsigned long flags);
+
+/** @} */
+
+
+
+/** \defgroup hwlocality_distances_remove Remove distances between objects
+ * @{
  */
-HWLOC_DECLSPEC int hwloc_distances_add(hwloc_topology_t topology,
-				       unsigned nbobjs, hwloc_obj_t *objs, hwloc_uint64_t *values,
-				       unsigned long kind, unsigned long flags);
 
 /** \brief Remove all distance matrices from a topology.
  *
@@ -258,18 +459,24 @@ HWLOC_DECLSPEC int hwloc_distances_add(hwloc_topology_t topology,
  *
  * If these distances were used to group objects, these additional
  * Group objects are not removed from the topology.
+ *
+ * \return 0 on success, -1 on error.
  */
 HWLOC_DECLSPEC int hwloc_distances_remove(hwloc_topology_t topology);
 
 /** \brief Remove distance matrices for objects at a specific depth in the topology.
  *
  * Identical to hwloc_distances_remove() but only applies to one level of the topology.
+ *
+ * \return 0 on success, -1 on error.
  */
 HWLOC_DECLSPEC int hwloc_distances_remove_by_depth(hwloc_topology_t topology, int depth);
 
 /** \brief Remove distance matrices for objects of a specific type in the topology.
  *
  * Identical to hwloc_distances_remove() but only applies to one level of the topology.
+ *
+ * \return 0 on success, -1 on error.
  */
 static __hwloc_inline int
 hwloc_distances_remove_by_type(hwloc_topology_t topology, hwloc_obj_type_t type)
@@ -283,6 +490,8 @@ hwloc_distances_remove_by_type(hwloc_topology_t topology, hwloc_obj_type_t type)
 /** \brief Release and remove the given distance matrice from the topology.
  *
  * This function includes a call to hwloc_distances_release().
+ *
+ * \return 0 on success, -1 on error.
  */
 HWLOC_DECLSPEC int hwloc_distances_release_remove(hwloc_topology_t topology, struct hwloc_distances_s *distances);
 

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

@@ -55,7 +55,7 @@ enum hwloc_topology_export_xml_flags_e {
  *
  * \p flags is a OR'ed set of ::hwloc_topology_export_xml_flags_e.
  *
- * \return -1 if a failure occured.
+ * \return 0 on success, or -1 on error.
  *
  * \note See also hwloc_topology_set_userdata_export_callback()
  * for exporting application-specific object userdata.
@@ -91,7 +91,7 @@ HWLOC_DECLSPEC int hwloc_topology_export_xml(hwloc_topology_t topology, const ch
  *
  * \p flags is a OR'ed set of ::hwloc_topology_export_xml_flags_e.
  *
- * \return -1 if a failure occured.
+ * \return 0 on success, or -1 on error.
  *
  * \note See also hwloc_topology_set_userdata_export_callback()
  * for exporting application-specific object userdata.
@@ -145,13 +145,15 @@ HWLOC_DECLSPEC void hwloc_topology_set_userdata_export_callback(hwloc_topology_t
  * that were given to the export callback.
  *
  * Only printable characters may be exported to XML string attributes.
- * If a non-printable character is passed in \p name or \p buffer,
- * the function returns -1 with errno set to EINVAL.
  *
  * If exporting binary data, the application should first encode into
  * printable characters only (or use hwloc_export_obj_userdata_base64()).
  * It should also take care of portability issues if the export may
  * be reimported on a different architecture.
+ *
+ * \return 0 on success.
+ * \return -1 with errno set to \c EINVAL if a non-printable character is
+ * passed in \p name or \b buffer.
  */
 HWLOC_DECLSPEC int hwloc_export_obj_userdata(void *reserved, hwloc_topology_t topology, hwloc_obj_t obj, const char *name, const void *buffer, size_t length);
 
@@ -165,8 +167,14 @@ HWLOC_DECLSPEC int hwloc_export_obj_userdata(void *reserved, hwloc_topology_t to
  * This function may only be called from within the export() callback passed
  * to hwloc_topology_set_userdata_export_callback().
  *
+ * The name must be made of printable characters for export to XML string attributes.
+ *
  * The function does not take care of portability issues if the export
  * may be reimported on a different architecture.
+ *
+ * \return 0 on success.
+ * \return -1 with errno set to \c EINVAL if a non-printable character is
+ * passed in \p name.
  */
 HWLOC_DECLSPEC int hwloc_export_obj_userdata_base64(void *reserved, hwloc_topology_t topology, hwloc_obj_t obj, const char *name, const void *buffer, size_t length);
 

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2012 Blue Brain Project, EPFL. All rights reserved.
- * Copyright © 2012-2013 Inria.  All rights reserved.
+ * Copyright © 2012-2023 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -39,9 +39,9 @@ extern "C" {
 /** \brief Get the hwloc OS device object corresponding to the
  * OpenGL display given by port and device index.
  *
- * Return the OS device object describing the OpenGL display
+ * \return The hwloc OS device object describing the OpenGL display
  * whose port (server) is \p port and device (screen) is \p device.
- * Return NULL if there is none.
+ * \return \c NULL if none could be found.
  *
  * The topology \p topology does not necessarily have to match the current
  * machine. For instance the topology may be an XML import of a remote host.
@@ -70,9 +70,9 @@ hwloc_gl_get_display_osdev_by_port_device(hwloc_topology_t topology,
 /** \brief Get the hwloc OS device object corresponding to the
  * OpenGL display given by name.
  *
- * Return the OS device object describing the OpenGL display
+ * \return The hwloc OS device object describing the OpenGL display
  * whose name is \p name, built as ":port.device" such as ":0.0" .
- * Return NULL if there is none.
+ * \return \c NULL if none could be found.
  *
  * The topology \p topology does not necessarily have to match the current
  * machine. For instance the topology may be an XML import of a remote host.
@@ -99,9 +99,11 @@ hwloc_gl_get_display_osdev_by_name(hwloc_topology_t topology,
 /** \brief Get the OpenGL display port and device corresponding
  * to the given hwloc OS object.
  *
- * Return the OpenGL display port (server) in \p port and device (screen)
+ * Retrieves the OpenGL display port (server) in \p port and device (screen)
  * in \p screen that correspond to the given hwloc OS device object.
- * Return \c -1 if there is none.
+ *
+ * \return 0 on success.
+ * \return -1 if none could be found.
  *
  * The topology \p topology does not necessarily have to match the current
  * machine. For instance the topology may be an XML import of a remote host.

src/3rdparty/hwloc/include/hwloc/glibc-sched.h

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2023 Inria.  All rights reserved.
  * Copyright © 2009-2011 Université Bordeaux
  * Copyright © 2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -52,6 +52,8 @@ extern "C" {
  * that takes a cpu_set_t as input parameter.
  *
  * \p schedsetsize should be sizeof(cpu_set_t) unless \p schedset was dynamically allocated with CPU_ALLOC
+ *
+ * \return 0.
  */
 static __hwloc_inline int
 hwloc_cpuset_to_glibc_sched_affinity(hwloc_topology_t topology __hwloc_attribute_unused, hwloc_const_cpuset_t hwlocset,
@@ -80,6 +82,9 @@ hwloc_cpuset_to_glibc_sched_affinity(hwloc_topology_t topology __hwloc_attribute
  * that takes a cpu_set_t  as input parameter.
  *
  * \p schedsetsize should be sizeof(cpu_set_t) unless \p schedset was dynamically allocated with CPU_ALLOC
+ *
+ * \return 0 on success.
+ * \return -1 with errno set to \c ENOMEM if some internal reallocation failed.
  */
 static __hwloc_inline int
 hwloc_cpuset_from_glibc_sched_affinity(hwloc_topology_t topology __hwloc_attribute_unused, hwloc_cpuset_t hwlocset,
@@ -95,7 +100,8 @@ hwloc_cpuset_from_glibc_sched_affinity(hwloc_topology_t topology __hwloc_attribu
   cpu = 0;
   while (count) {
     if (CPU_ISSET_S(cpu, schedsetsize, schedset)) {
-      hwloc_bitmap_set(hwlocset, cpu);
+      if (hwloc_bitmap_set(hwlocset, cpu) < 0)
+        return -1;
       count--;
     }
     cpu++;
@@ -107,7 +113,8 @@ hwloc_cpuset_from_glibc_sched_affinity(hwloc_topology_t topology __hwloc_attribu
   assert(schedsetsize == sizeof(cpu_set_t));
   for(cpu=0; cpu<CPU_SETSIZE; cpu++)
     if (CPU_ISSET(cpu, schedset))
-      hwloc_bitmap_set(hwlocset, cpu);
+      if (hwloc_bitmap_set(hwlocset, cpu) < 0)
+        return -1;
 #endif /* !CPU_ZERO_S */
   return 0;
 }

src/3rdparty/hwloc/include/hwloc/intel-mic.h

@@ -1,136 +0,0 @@
-/*
- * Copyright © 2013-2016 Inria.  All rights reserved.
- * See COPYING in top-level directory.
- */
-
-/** \file
- * \brief Macros to help interaction between hwloc and Intel Xeon Phi (MIC).
- *
- * Applications that use both hwloc and Intel Xeon Phi (MIC) may want to
- * include this file so as to get topology information for MIC devices.
- */
-
-#ifndef HWLOC_INTEL_MIC_H
-#define HWLOC_INTEL_MIC_H
-
-#include "hwloc.h"
-#include "hwloc/autogen/config.h"
-#include "hwloc/helper.h"
-
-#ifdef HWLOC_LINUX_SYS
-#include "hwloc/linux.h"
-
-#include <dirent.h>
-#include <string.h>
-#endif
-
-#include <stdio.h>
-#include <stdlib.h>
-
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-
-/** \defgroup hwlocality_intel_mic Interoperability with Intel Xeon Phi (MIC)
- *
- * This interface offers ways to retrieve topology information about
- * Intel Xeon Phi (MIC) devices.
- *
- * @{
- */
-
-/** \brief Get the CPU set of logical processors that are physically
- * close to MIC device whose index is \p idx.
- *
- * Return the CPU set describing the locality of the MIC device whose index is \p idx.
- *
- * Topology \p topology and device index \p idx must match the local machine.
- * I/O devices detection is not needed in the topology.
- *
- * The function only returns the locality of the device.
- * If more information about the device is needed, OS objects should
- * be used instead, see hwloc_intel_mic_get_device_osdev_by_index().
- *
- * This function is currently only implemented in a meaningful way for
- * Linux; other systems will simply get a full cpuset.
- */
-static __hwloc_inline int
-hwloc_intel_mic_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
-				  int idx __hwloc_attribute_unused,
-				  hwloc_cpuset_t set)
-{
-#ifdef HWLOC_LINUX_SYS
-	/* If we're on Linux, use the sysfs mechanism to get the local cpus */
-#define HWLOC_INTEL_MIC_DEVICE_SYSFS_PATH_MAX 128
-	char path[HWLOC_INTEL_MIC_DEVICE_SYSFS_PATH_MAX];
-	DIR *sysdir = NULL;
-	struct dirent *dirent;
-	unsigned pcibus, pcidev, pcifunc;
-
-	if (!hwloc_topology_is_thissystem(topology)) {
-		errno = EINVAL;
-		return -1;
-	}
-
-	sprintf(path, "/sys/class/mic/mic%d", idx);
-	sysdir = opendir(path);
-	if (!sysdir)
-		return -1;
-
-	while ((dirent = readdir(sysdir)) != NULL) {
-		if (sscanf(dirent->d_name, "pci_%02x:%02x.%02x", &pcibus, &pcidev, &pcifunc) == 3) {
-			sprintf(path, "/sys/class/mic/mic%d/pci_%02x:%02x.%02x/local_cpus", idx, pcibus, pcidev, pcifunc);
-			if (hwloc_linux_read_path_as_cpumask(path, set) < 0
-			    || hwloc_bitmap_iszero(set))
-				hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
-			break;
-		}
-	}
-
-	closedir(sysdir);
-#else
-	/* Non-Linux systems simply get a full cpuset */
-	hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
-#endif
-	return 0;
-}
-
-/** \brief Get the hwloc OS device object corresponding to the
- * MIC device for the given index.
- *
- * Return the OS device object describing the MIC device whose index is \p idx.
- * Return NULL if there is none.
- *
- * The topology \p topology does not necessarily have to match the current
- * machine. For instance the topology may be an XML import of a remote host.
- * I/O devices detection must be enabled in the topology.
- *
- * \note The corresponding PCI device object can be obtained by looking
- * at the OS device parent object.
- */
-static __hwloc_inline hwloc_obj_t
-hwloc_intel_mic_get_device_osdev_by_index(hwloc_topology_t topology,
-					  unsigned idx)
-{
-	hwloc_obj_t osdev = NULL;
-	while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
-		if (HWLOC_OBJ_OSDEV_COPROC == osdev->attr->osdev.type
-                    && osdev->name
-		    && !strncmp("mic", osdev->name, 3)
-		    && atoi(osdev->name + 3) == (int) idx)
-                        return osdev;
-        }
-        return NULL;
-}
-
-/** @} */
-
-
-#ifdef __cplusplus
-} /* extern "C" */
-#endif
-
-
-#endif /* HWLOC_INTEL_MIC_H */

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

@@ -0,0 +1,298 @@
+/*
+ * Copyright © 2021-2024 Inria.  All rights reserved.
+ * See COPYING in top-level directory.
+ */
+
+/** \file
+ * \brief Macros to help interaction between hwloc and the oneAPI Level Zero interface.
+ *
+ * Applications that use both hwloc and Level Zero may want to
+ * include this file so as to get topology information for L0 devices.
+ */
+
+#ifndef HWLOC_LEVELZERO_H
+#define HWLOC_LEVELZERO_H
+
+#include "hwloc.h"
+#include "hwloc/autogen/config.h"
+#include "hwloc/helper.h"
+#ifdef HWLOC_LINUX_SYS
+#include "hwloc/linux.h"
+#endif
+
+#include <level_zero/ze_api.h>
+#include <level_zero/zes_api.h>
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+/** \defgroup hwlocality_levelzero Interoperability with the oneAPI Level Zero interface.
+ *
+ * This interface offers ways to retrieve topology information about
+ * devices managed by the Level Zero API, both for main Core devices (ZE API)
+ * and the Sysman devices (ZES API).
+ *
+ * @{
+ */
+
+/** \brief Get the CPU set of logical processors that are physically
+ * close to the Level Zero device \p device
+ *
+ * Store in \p set the CPU-set describing the locality of
+ * the Level Zero device \p device.
+ *
+ * Topology \p topology and device \p device must match the local machine.
+ * The Level Zero library must have been initialized with zeInit().
+ * I/O devices detection and the Level Zero component are not needed in the
+ * topology.
+ *
+ * The function only returns the locality of the device.
+ * If more information about the device is needed, OS objects should
+ * be used instead, see hwloc_levelzero_get_device_osdev().
+ *
+ * This function is currently only implemented in a meaningful way for
+ * Linux; other systems will simply get a full cpuset.
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if device information could not be found.
+ *
+ * \note zeDevicePciGetPropertiesExt() must be supported, or the entire machine
+ * locality will be returned.
+ */
+static __hwloc_inline int
+hwloc_levelzero_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
+                                  ze_device_handle_t device, hwloc_cpuset_t set)
+{
+#ifdef HWLOC_LINUX_SYS
+  /* If we're on Linux, use the sysfs mechanism to get the local cpus */
+#define HWLOC_LEVELZERO_DEVICE_SYSFS_PATH_MAX 128
+  char path[HWLOC_LEVELZERO_DEVICE_SYSFS_PATH_MAX];
+  ze_pci_ext_properties_t pci;
+  ze_result_t res;
+
+  if (!hwloc_topology_is_thissystem(topology)) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  pci.stype =  ZE_STRUCTURE_TYPE_PCI_EXT_PROPERTIES;
+  pci.pNext = NULL;
+  res = zeDevicePciGetPropertiesExt(device, &pci);
+  if (res != ZE_RESULT_SUCCESS) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.%01x/local_cpus",
+          pci.address.domain, pci.address.bus, pci.address.device, pci.address.function);
+  if (hwloc_linux_read_path_as_cpumask(path, set) < 0
+      || hwloc_bitmap_iszero(set))
+    hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
+#else
+  /* Non-Linux systems simply get a full cpuset */
+  hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
+#endif
+  return 0;
+}
+
+/** \brief Get the CPU set of logical processors that are physically
+ * close to the Level Zero Sysman device \p device
+ *
+ * Store in \p set the CPU-set describing the locality of
+ * the Level Zero device \p device.
+ *
+ * Topology \p topology and device \p device must match the local machine.
+ * The Level Zero library must have been initialized with Sysman enabled
+ * with zesInit().
+ * I/O devices detection and the Level Zero component are not needed in the
+ * topology.
+ *
+ * The function only returns the locality of the device.
+ * If more information about the device is needed, OS objects should
+ * be used instead, see hwloc_levelzero_get_device_osdev().
+ *
+ * This function is currently only implemented in a meaningful way for
+ * Linux; other systems will simply get a full cpuset.
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if device information could not be found.
+ */
+static __hwloc_inline int
+hwloc_levelzero_get_sysman_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
+                                         zes_device_handle_t device, hwloc_cpuset_t set)
+{
+#ifdef HWLOC_LINUX_SYS
+  /* If we're on Linux, use the sysfs mechanism to get the local cpus */
+#define HWLOC_LEVELZERO_DEVICE_SYSFS_PATH_MAX 128
+  char path[HWLOC_LEVELZERO_DEVICE_SYSFS_PATH_MAX];
+  zes_pci_properties_t pci;
+  ze_result_t res;
+
+  if (!hwloc_topology_is_thissystem(topology)) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  res = zesDevicePciGetProperties(device, &pci);
+  if (res != ZE_RESULT_SUCCESS) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.%01x/local_cpus",
+          pci.address.domain, pci.address.bus, pci.address.device, pci.address.function);
+  if (hwloc_linux_read_path_as_cpumask(path, set) < 0
+      || hwloc_bitmap_iszero(set))
+    hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
+#else
+  /* Non-Linux systems simply get a full cpuset */
+  hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
+#endif
+  return 0;
+}
+
+/** \brief Get the hwloc OS device object corresponding to Level Zero device
+ * \p device.
+ *
+ * \return The hwloc OS device object that describes the given Level Zero device \p device.
+ * \return \c NULL if none could be found.
+ *
+ * Topology \p topology and device \p dv_ind must match the local machine.
+ * The Level Zero library must have been initialized with zeInit().
+ * I/O devices detection and the Level Zero component must be enabled in the
+ * topology. If not, the locality of the object may still be found using
+ * hwloc_levelzero_get_device_cpuset().
+ *
+ * \note If the input ZE device is actually a subdevice, then its parent
+ * (root device) is actually translated, i.e. the main hwloc OS device
+ * is returned instead of one of its children.
+ *
+ * \note The corresponding hwloc PCI device may be found by looking
+ * at the result parent pointer (unless PCI devices are filtered out).
+ *
+ * \note zeDevicePciGetPropertiesExt() must be supported.
+ */
+static __hwloc_inline hwloc_obj_t
+hwloc_levelzero_get_device_osdev(hwloc_topology_t topology, ze_device_handle_t device)
+{
+  ze_pci_ext_properties_t pci;
+  ze_result_t res;
+  hwloc_obj_t osdev;
+
+  if (!hwloc_topology_is_thissystem(topology)) {
+    errno = EINVAL;
+    return NULL;
+  }
+
+  pci.stype = ZE_STRUCTURE_TYPE_PCI_EXT_PROPERTIES;
+  pci.pNext = NULL;
+  res = zeDevicePciGetPropertiesExt(device, &pci);
+  if (res != ZE_RESULT_SUCCESS) {
+    errno = EINVAL;
+    return NULL;
+  }
+
+  osdev = NULL;
+  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
+    hwloc_obj_t pcidev;
+
+    if (strncmp(osdev->name, "ze", 2))
+      continue;
+
+    pcidev = osdev;
+    while (pcidev && pcidev->type != HWLOC_OBJ_PCI_DEVICE)
+      pcidev = pcidev->parent;
+    if (!pcidev)
+      continue;
+
+    if (pcidev
+      && pcidev->type == HWLOC_OBJ_PCI_DEVICE
+      && pcidev->attr->pcidev.domain == pci.address.domain
+      && pcidev->attr->pcidev.bus == pci.address.bus
+      && pcidev->attr->pcidev.dev == pci.address.device
+      && pcidev->attr->pcidev.func == pci.address.function)
+      return osdev;
+
+    /* FIXME: when we'll have serialnumber, try it in case PCI is filtered-out */
+  }
+
+  return NULL;
+}
+
+/** \brief Get the hwloc OS device object corresponding to Level Zero Sysman device
+ * \p device.
+ *
+ * \return The hwloc OS device object that describes the given Level Zero device \p device.
+ * \return \c NULL if none could be found.
+ *
+ * Topology \p topology and device \p dv_ind must match the local machine.
+ * The Level Zero library must have been initialized with Sysman enabled
+ * with zesInit().
+ * I/O devices detection and the Level Zero component must be enabled in the
+ * topology. If not, the locality of the object may still be found using
+ * hwloc_levelzero_get_device_cpuset().
+ *
+ * \note If the input ZES device is actually a subdevice, then its parent
+ * (root device) is actually translated, i.e. the main hwloc OS device
+ * is returned instead of one of its children.
+ *
+ * \note The corresponding hwloc PCI device may be found by looking
+ * at the result parent pointer (unless PCI devices are filtered out).
+ */
+static __hwloc_inline hwloc_obj_t
+hwloc_levelzero_get_sysman_device_osdev(hwloc_topology_t topology, zes_device_handle_t device)
+{
+  zes_pci_properties_t pci;
+  ze_result_t res;
+  hwloc_obj_t osdev;
+
+  if (!hwloc_topology_is_thissystem(topology)) {
+    errno = EINVAL;
+    return NULL;
+  }
+
+  res = zesDevicePciGetProperties(device, &pci);
+  if (res != ZE_RESULT_SUCCESS) {
+    errno = EINVAL;
+    return NULL;
+  }
+
+  osdev = NULL;
+  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
+    hwloc_obj_t pcidev;
+
+    if (strncmp(osdev->name, "ze", 2))
+      continue;
+
+    pcidev = osdev;
+    while (pcidev && pcidev->type != HWLOC_OBJ_PCI_DEVICE)
+      pcidev = pcidev->parent;
+    if (!pcidev)
+      continue;
+
+    if (pcidev
+      && pcidev->type == HWLOC_OBJ_PCI_DEVICE
+      && pcidev->attr->pcidev.domain == pci.address.domain
+      && pcidev->attr->pcidev.bus == pci.address.bus
+      && pcidev->attr->pcidev.dev == pci.address.device
+      && pcidev->attr->pcidev.func == pci.address.function)
+      return osdev;
+
+    /* FIXME: when we'll have serialnumber, try it in case PCI is filtered-out */
+  }
+
+  return NULL;
+}
+
+/** @} */
+
+
+#ifdef __cplusplus
+} /* extern "C" */
+#endif
+
+
+#endif /* HWLOC_LEVELZERO_H */

src/3rdparty/hwloc/include/hwloc/linux-libnuma.h

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2017 Inria.  All rights reserved.
+ * Copyright © 2009-2023 Inria.  All rights reserved.
  * Copyright © 2009-2010, 2012 Université Bordeaux
  * See COPYING in top-level directory.
  */
@@ -50,6 +50,8 @@ extern "C" {
  * This function may be used before calling set_mempolicy, mbind, migrate_pages
  * or any other function that takes an array of unsigned long and a maximal
  * node number as input parameter.
+ *
+ * \return 0.
  */
 static __hwloc_inline int
 hwloc_cpuset_to_linux_libnuma_ulongs(hwloc_topology_t topology, hwloc_const_cpuset_t cpuset,
@@ -84,6 +86,8 @@ hwloc_cpuset_to_linux_libnuma_ulongs(hwloc_topology_t topology, hwloc_const_cpus
  * This function may be used before calling set_mempolicy, mbind, migrate_pages
  * or any other function that takes an array of unsigned long and a maximal
  * node number as input parameter.
+ *
+ * \return 0.
  */
 static __hwloc_inline int
 hwloc_nodeset_to_linux_libnuma_ulongs(hwloc_topology_t topology, hwloc_const_nodeset_t nodeset,
@@ -119,6 +123,9 @@ hwloc_nodeset_to_linux_libnuma_ulongs(hwloc_topology_t topology, hwloc_const_nod
  * This function may be used after calling get_mempolicy or any other function
  * that takes an array of unsigned long as output parameter (and possibly
  * a maximal node number as input parameter).
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if failing an internal reallocation.
  */
 static __hwloc_inline int
 hwloc_cpuset_from_linux_libnuma_ulongs(hwloc_topology_t topology, hwloc_cpuset_t cpuset,
@@ -130,7 +137,8 @@ hwloc_cpuset_from_linux_libnuma_ulongs(hwloc_topology_t topology, hwloc_cpuset_t
   while ((node = hwloc_get_next_obj_by_depth(topology, depth, node)) != NULL)
     if (node->os_index < maxnode
 	&& (mask[node->os_index/sizeof(*mask)/8] & (1UL << (node->os_index % (sizeof(*mask)*8)))))
-      hwloc_bitmap_or(cpuset, cpuset, node->cpuset);
+      if (hwloc_bitmap_or(cpuset, cpuset, node->cpuset) < 0)
+        return -1;
   return 0;
 }
 
@@ -142,6 +150,9 @@ hwloc_cpuset_from_linux_libnuma_ulongs(hwloc_topology_t topology, hwloc_cpuset_t
  * This function may be used after calling get_mempolicy or any other function
  * that takes an array of unsigned long as output parameter (and possibly
  * a maximal node number as input parameter).
+ *
+ * \return 0 on success.
+ * \return -1 with errno set to \c ENOMEM if some internal reallocation failed.
  */
 static __hwloc_inline int
 hwloc_nodeset_from_linux_libnuma_ulongs(hwloc_topology_t topology, hwloc_nodeset_t nodeset,
@@ -153,7 +164,8 @@ hwloc_nodeset_from_linux_libnuma_ulongs(hwloc_topology_t topology, hwloc_nodeset
   while ((node = hwloc_get_next_obj_by_depth(topology, depth, node)) != NULL)
     if (node->os_index < maxnode
 	&& (mask[node->os_index/sizeof(*mask)/8] & (1UL << (node->os_index % (sizeof(*mask)*8)))))
-      hwloc_bitmap_set(nodeset, node->os_index);
+      if (hwloc_bitmap_set(nodeset, node->os_index) < 0)
+        return -1;
   return 0;
 }
 
@@ -184,7 +196,7 @@ hwloc_nodeset_from_linux_libnuma_ulongs(hwloc_topology_t topology, hwloc_nodeset
  * This function may be used before calling many numa_ functions
  * that use a struct bitmask as an input parameter.
  *
- * \return newly allocated struct bitmask.
+ * \return newly allocated struct bitmask, or \c NULL on error.
  */
 static __hwloc_inline struct bitmask *
 hwloc_cpuset_to_linux_libnuma_bitmask(hwloc_topology_t topology, hwloc_const_cpuset_t cpuset) __hwloc_attribute_malloc;
@@ -209,7 +221,7 @@ hwloc_cpuset_to_linux_libnuma_bitmask(hwloc_topology_t topology, hwloc_const_cpu
  * This function may be used before calling many numa_ functions
  * that use a struct bitmask as an input parameter.
  *
- * \return newly allocated struct bitmask.
+ * \return newly allocated struct bitmask, or \c NULL on error.
  */
 static __hwloc_inline struct bitmask *
 hwloc_nodeset_to_linux_libnuma_bitmask(hwloc_topology_t topology, hwloc_const_nodeset_t nodeset) __hwloc_attribute_malloc;
@@ -231,6 +243,9 @@ hwloc_nodeset_to_linux_libnuma_bitmask(hwloc_topology_t topology, hwloc_const_no
  *
  * This function may be used after calling many numa_ functions
  * that use a struct bitmask as an output parameter.
+ *
+ * \return 0 on success.
+ * \return -1 with errno set to \c ENOMEM if some internal reallocation failed.
  */
 static __hwloc_inline int
 hwloc_cpuset_from_linux_libnuma_bitmask(hwloc_topology_t topology, hwloc_cpuset_t cpuset,
@@ -241,7 +256,8 @@ hwloc_cpuset_from_linux_libnuma_bitmask(hwloc_topology_t topology, hwloc_cpuset_
   hwloc_bitmap_zero(cpuset);
   while ((node = hwloc_get_next_obj_by_depth(topology, depth, node)) != NULL)
     if (numa_bitmask_isbitset(bitmask, node->os_index))
-      hwloc_bitmap_or(cpuset, cpuset, node->cpuset);
+      if (hwloc_bitmap_or(cpuset, cpuset, node->cpuset) < 0)
+        return -1;
   return 0;
 }
 
@@ -249,6 +265,9 @@ hwloc_cpuset_from_linux_libnuma_bitmask(hwloc_topology_t topology, hwloc_cpuset_
  *
  * This function may be used after calling many numa_ functions
  * that use a struct bitmask as an output parameter.
+ *
+ * \return 0 on success.
+ * \return -1 with errno set to \c ENOMEM if some internal reallocation failed.
  */
 static __hwloc_inline int
 hwloc_nodeset_from_linux_libnuma_bitmask(hwloc_topology_t topology, hwloc_nodeset_t nodeset,
@@ -259,7 +278,8 @@ hwloc_nodeset_from_linux_libnuma_bitmask(hwloc_topology_t topology, hwloc_nodese
   hwloc_bitmap_zero(nodeset);
   while ((node = hwloc_get_next_obj_by_depth(topology, depth, node)) != NULL)
     if (numa_bitmask_isbitset(bitmask, node->os_index))
-      hwloc_bitmap_set(nodeset, node->os_index);
+      if (hwloc_bitmap_set(nodeset, node->os_index) < 0)
+        return -1;
   return 0;
 }
 

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2016 Inria.  All rights reserved.
+ * Copyright © 2009-2023 Inria.  All rights reserved.
  * Copyright © 2009-2011 Université Bordeaux
  * See COPYING in top-level directory.
  */
@@ -38,22 +38,35 @@ extern "C" {
  * The behavior is exactly the same as the Linux sched_setaffinity system call,
  * but uses a hwloc cpuset.
  *
+ * \return 0 on success, -1 on error.
+ *
  * \note This is equivalent to calling hwloc_set_proc_cpubind() with
  * HWLOC_CPUBIND_THREAD as flags.
  */
 HWLOC_DECLSPEC int hwloc_linux_set_tid_cpubind(hwloc_topology_t topology, pid_t tid, hwloc_const_cpuset_t set);
 
 /** \brief Get the current binding of thread \p tid
+ *
+ * The CPU-set \p set (previously allocated by the caller)
+ * is filled with the list of PUs which the thread
+ * was last bound to.
  *
  * The behavior is exactly the same as the Linux sched_getaffinity system call,
  * but uses a hwloc cpuset.
  *
+ * \return 0 on success, -1 on error.
+ *
  * \note This is equivalent to calling hwloc_get_proc_cpubind() with
  * ::HWLOC_CPUBIND_THREAD as flags.
  */
 HWLOC_DECLSPEC int hwloc_linux_get_tid_cpubind(hwloc_topology_t topology, pid_t tid, hwloc_cpuset_t set);
 
 /** \brief Get the last physical CPU where thread \p tid ran.
+ *
+ * The CPU-set \p set (previously allocated by the caller)
+ * is filled with the PU which the thread last ran on.
+ *
+ * \return 0 on success, -1 on error.
  *
  * \note This is equivalent to calling hwloc_get_proc_last_cpu_location() with
  * ::HWLOC_CPUBIND_THREAD as flags.
@@ -65,6 +78,8 @@ HWLOC_DECLSPEC int hwloc_linux_get_tid_last_cpu_location(hwloc_topology_t topolo
  * Might be used when reading CPU set from sysfs attributes such as topology
  * and caches for processors, or local_cpus for devices.
  *
+ * \return 0 on success, -1 on error.
+ *
  * \note This function ignores the HWLOC_FSROOT environment variable.
  */
 HWLOC_DECLSPEC int hwloc_linux_read_path_as_cpumask(const char *path, hwloc_bitmap_t set);

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

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2019-2020 Inria.  All rights reserved.
+ * Copyright © 2019-2025 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -54,6 +54,17 @@ extern "C" {
  * Attribute values for these nodes, if any, may then be obtained with
  * hwloc_memattr_get_value() and manually compared with the desired criteria.
  *
+ * Memory attributes are also used internally to build Memory Tiers which provide
+ * an easy way to distinguish NUMA nodes of different kinds, as explained
+ * in \ref heteromem.
+ *
+ * Beside tiers, hwloc defines a set of "default" nodes where normal memory
+ * allocations should be made from (see hwloc_topology_get_default_nodeset()).
+ * This is also useful for dividing the machine into a set of non-overlapping
+ * NUMA domains, for instance for binding tasks per domain.
+ *
+ * \sa An example is available in doc/examples/memory-attributes.c in the source tree.
+ *
  * \note The API also supports specific objects as initiator,
  * but it is currently not used internally by hwloc.
  * Users may for instance use it to provide custom performance
@@ -63,21 +74,26 @@ extern "C" {
  * @{
  */
 
-/** \brief Memory node attributes. */
+/** \brief Predefined memory attribute IDs.
+ * See ::hwloc_memattr_id_t for the generic definition of IDs
+ * for predefined or custom attributes.
+ */
 enum hwloc_memattr_id_e {
-  /** \brief "Capacity".
+  /** \brief
-   * The capacity is returned in bytes
+   * The \"Capacity\" is returned in bytes (local_memory attribute in objects).
-   * (local_memory attribute in objects).
    *
    * Best capacity nodes are nodes with <b>higher capacity</b>.
    *
    * No initiator is involved when looking at this attribute.
    * The corresponding attribute flags are ::HWLOC_MEMATTR_FLAG_HIGHER_FIRST.
+   *
+   * Capacity values may not be modified using hwloc_memattr_set_value().
+   * \hideinitializer
    */
   HWLOC_MEMATTR_ID_CAPACITY = 0,
 
-  /** \brief "Locality".
+  /** \brief
-   * The locality is returned as the number of PUs in that locality
+   * The \"Locality\" is returned as the number of PUs in that locality
    * (e.g. the weight of its cpuset).
    *
    * Best locality nodes are nodes with <b>smaller locality</b>
@@ -87,34 +103,108 @@ enum hwloc_memattr_id_e {
    *
    * No initiator is involved when looking at this attribute.
    * The corresponding attribute flags are ::HWLOC_MEMATTR_FLAG_HIGHER_FIRST.
+
+   * Locality values may not be modified using hwloc_memattr_set_value().
+   * \hideinitializer
    */
   HWLOC_MEMATTR_ID_LOCALITY = 1,
 
-  /** \brief "Bandwidth".
+  /** \brief
-   * The bandwidth is returned in MiB/s, as seen from the given initiator location.
+   * The \"Bandwidth\" is returned in MiB/s, as seen from the given initiator location.
+   *
    * Best bandwidth nodes are nodes with <b>higher bandwidth</b>.
+   *
    * The corresponding attribute flags are ::HWLOC_MEMATTR_FLAG_HIGHER_FIRST
    * and ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR.
+   *
+   * This is the average bandwidth for read and write accesses. If the platform
+   * provides individual read and write bandwidths but no explicit average value,
+   * hwloc computes and returns the average.
+   * \hideinitializer
    */
   HWLOC_MEMATTR_ID_BANDWIDTH = 2,
 
-  /** \brief "Latency".
+  /** \brief
-   * The latency is returned as nanoseconds, as seen from the given initiator location.
+   * The \"ReadBandwidth\" is returned in MiB/s, as seen from the given initiator location.
+   *
+   * Best bandwidth nodes are nodes with <b>higher bandwidth</b>.
+   *
+   * The corresponding attribute flags are ::HWLOC_MEMATTR_FLAG_HIGHER_FIRST
+   * and ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR.
+   * \hideinitializer
+   */
+  HWLOC_MEMATTR_ID_READ_BANDWIDTH = 4,
+
+  /** \brief
+   * The \"WriteBandwidth\" is returned in MiB/s, as seen from the given initiator location.
+   *
+   * Best bandwidth nodes are nodes with <b>higher bandwidth</b>.
+   *
+   * The corresponding attribute flags are ::HWLOC_MEMATTR_FLAG_HIGHER_FIRST
+   * and ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR.
+   * \hideinitializer
+   */
+  HWLOC_MEMATTR_ID_WRITE_BANDWIDTH = 5,
+
+  /** \brief
+   * The \"Latency\" is returned as nanoseconds, as seen from the given initiator location.
+   *
    * Best latency nodes are nodes with <b>smaller latency</b>.
+   *
    * The corresponding attribute flags are ::HWLOC_MEMATTR_FLAG_LOWER_FIRST
    * and ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR.
+   *
+   * This is the average latency for read and write accesses. If the platform
+   * provides individual read and write latencies but no explicit average value,
+   * hwloc computes and returns the average.
+   * \hideinitializer
    */
-  HWLOC_MEMATTR_ID_LATENCY = 3
+  HWLOC_MEMATTR_ID_LATENCY = 3,
 
-  /* TODO read vs write, persistence? */
+  /** \brief
+   * The \"ReadLatency\" is returned as nanoseconds, as seen from the given initiator location.
+   *
+   * Best latency nodes are nodes with <b>smaller latency</b>.
+   *
+   * The corresponding attribute flags are ::HWLOC_MEMATTR_FLAG_LOWER_FIRST
+   * and ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR.
+   * \hideinitializer
+   */
+  HWLOC_MEMATTR_ID_READ_LATENCY = 6,
+
+  /** \brief
+   * The \"WriteLatency\" is returned as nanoseconds, as seen from the given initiator location.
+   *
+   * Best latency nodes are nodes with <b>smaller latency</b>.
+   *
+   * The corresponding attribute flags are ::HWLOC_MEMATTR_FLAG_LOWER_FIRST
+   * and ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR.
+   * \hideinitializer
+   */
+  HWLOC_MEMATTR_ID_WRITE_LATENCY = 7,
+
+  /* TODO persistence? */
+
+  HWLOC_MEMATTR_ID_MAX /**< \private
+                        * Sentinel value for predefined attributes.
+                        * Dynamically registered custom attributes start here.
+                        */
 };
 
 /** \brief A memory attribute identifier.
- * May be either one of ::hwloc_memattr_id_e or a new id returned by hwloc_memattr_register().
+ *
+ * hwloc predefines some commonly-used attributes in ::hwloc_memattr_id_e.
+ * One may then dynamically register custom ones with hwloc_memattr_register(),
+ * they will be assigned IDs immediately after the predefined ones.
+ * See \ref hwlocality_memattrs_manage for more information about
+ * existing attribute IDs.
  */
 typedef unsigned hwloc_memattr_id_t;
 
 /** \brief Return the identifier of the memory attribute with the given name.
+ *
+ * \return 0 on success.
+ * \return -1 with errno set to \c EINVAL if no such attribute exists.
  */
 HWLOC_DECLSPEC int
 hwloc_memattr_get_by_name(hwloc_topology_t topology,
@@ -160,6 +250,16 @@ enum hwloc_local_numanode_flag_e {
    */
   HWLOC_LOCAL_NUMANODE_FLAG_SMALLER_LOCALITY = (1UL<<1),
 
+  /** \breif Select NUMA nodes whose locality intersects the given cpuset.
+   * This includes larger and smaller localities as well as localities
+   * that are partially included.
+   * For instance, if the locality is one core of both packages, a NUMA node
+   * local to one package is neither larger nor smaller than this locality,
+   * but it intersects it.
+   * \hideinitializer
+   */
+  HWLOC_LOCAL_NUMANODE_FLAG_INTERSECT_LOCALITY = (1UL<<3),
+
   /** \brief Select all NUMA nodes in the topology.
    * The initiator \p initiator is ignored.
    * \hideinitializer
@@ -184,6 +284,8 @@ enum hwloc_local_numanode_flag_e {
  * or the number of nodes that would have been stored if there were
  * enough room.
  *
+ * \return 0 on success or -1 on error.
+ *
  * \note Some of these NUMA nodes may not have any memory attribute
  * values and hence not be reported as actual targets in other functions.
  *
@@ -203,7 +305,57 @@ hwloc_get_local_numanode_objs(hwloc_topology_t topology,
                               hwloc_obj_t *nodes,
                               unsigned long flags);
 
-
+/** \brief Return the set of default NUMA nodes
+ *
+ * In machines with heterogeneous memory, some NUMA nodes are considered
+ * the default ones, i.e. where basic allocations should be made from.
+ * These are usually DRAM nodes.
+ *
+ * Other nodes may be reserved for specific use (I/O device memory, e.g. GPU memory),
+ * small but high performance (HBM), large but slow memory (NVM), etc.
+ * Buffers should usually not be allocated from there unless explicitly required.
+ *
+ * This function fills \p nodeset with the bits of NUMA nodes considered default.
+ *
+ * It is guaranteed that these nodes have non-intersecting CPU sets,
+ * i.e. cores may not have multiple local NUMA nodes anymore.
+ * Hence this may be used to iterate over the platform divided into separate
+ * NUMA localities, for instance for binding one task per NUMA domain.
+ *
+ * Any core that had some local NUMA node(s) in the initial topology should
+ * still have one in the default nodeset. Corner cases where this would be
+ * wrong consist in asymmetric platforms with missing DRAM nodes, or topologies
+ * that were already restricted to less NUMA nodes.
+ *
+ * The returned nodeset may be passed to hwloc_topology_restrict() with
+ * ::HWLOC_RESTRICT_FLAG_BYNODESET to remove all non-default nodes from
+ * the topology. The resulting topology will be easier to use when iterating
+ * over (now homogeneous) NUMA nodes.
+ *
+ * The heuristics for finding default nodes relies on memory tiers and subtypes
+ * (see \ref heteromem) as well as the assumption that hardware vendors list
+ * default nodes first in hardware tables.
+ *
+ * \p flags must be \c 0 for now.
+ *
+ * \return 0 on success.
+ * \return -1 on error.
+ *
+ * \note The returned nodeset usually contains all nodes from a single memory
+ * tier, likely the DRAM one.
+ *
+ * \note The returned nodeset is included in the list of available nodes
+ * returned by hwloc_topology_get_topology_nodeset(). It is strictly smaller
+ * if the machine has heterogeneous memory.
+ *
+ * \note The heuristics may return a suboptimal set of nodes if hwloc could
+ * not guess memory types and/or if some default nodes were removed earlier
+ * from the topology (e.g. with hwloc_topology_restrict()).
+ */
+HWLOC_DECLSPEC int
+hwloc_topology_get_default_nodeset(hwloc_topology_t topology,
+                                   hwloc_nodeset_t nodeset,
+                                   unsigned long flags);
 
 /** \brief Return an attribute value for a specific target NUMA node.
  *
@@ -211,8 +363,16 @@ hwloc_get_local_numanode_objs(hwloc_topology_t topology,
  * (it does not have the flag ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR),
  * location \p initiator is ignored and may be \c NULL.
  *
+ * \p target_node cannot be \c NULL. If \p attribute is ::HWLOC_MEMATTR_ID_CAPACITY,
+ * \p target_node must be a NUMA node. If it is ::HWLOC_MEMATTR_ID_LOCALITY,
+ * \p target_node must have a CPU set.
+ *
  * \p flags must be \c 0 for now.
  *
+ * \return 0 on success.
+ * \return -1 on error, for instance with errno set to \c EINVAL if flags
+ * are invalid or no such attribute exists.
+ *
  * \note The initiator \p initiator should be of type ::HWLOC_LOCATION_TYPE_CPUSET
  * when refering to accesses performed by CPU cores.
  * ::HWLOC_LOCATION_TYPE_OBJECT is currently unused internally by hwloc,
@@ -244,7 +404,10 @@ hwloc_memattr_get_value(hwloc_topology_t topology,
  *
  * \p flags must be \c 0 for now.
  *
- * If there are no matching targets, \c -1 is returned with \p errno set to \c ENOENT;
+ * \return 0 on success.
+ * \return -1 with errno set to \c ENOENT if there are no matching targets.
+ * \return -1 with errno set to \c EINVAL if flags are invalid,
+ * or no such attribute exists.
  *
  * \note The initiator \p initiator should be of type ::HWLOC_LOCATION_TYPE_CPUSET
  * when refering to accesses performed by CPU cores.
@@ -260,10 +423,6 @@ hwloc_memattr_get_best_target(hwloc_topology_t topology,
                               hwloc_obj_t *best_target, hwloc_uint64_t *value);
 
 /** \brief Return the best initiator for the given attribute and target NUMA node.
- *
- * If the attribute does not relate to a specific initiator
- * (it does not have the flag ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR),
- * \c -1 is returned and \p errno is set to \c EINVAL.
  *
  * If \p value is non \c NULL, the corresponding value is returned there.
  *
@@ -277,96 +436,22 @@ hwloc_memattr_get_best_target(hwloc_topology_t topology,
  * The returned initiator should not be modified or freed,
  * it belongs to the topology.
  *
+ * \p target_node cannot be \c NULL.
+ *
  * \p flags must be \c 0 for now.
  *
- * If there are no matching initiators, \c -1 is returned with \p errno set to \c ENOENT;
+ * \return 0 on success.
+ * \return -1 with errno set to \c ENOENT if there are no matching initiators.
+ * \return -1 with errno set to \c EINVAL if the attribute does not relate to a specific initiator
+ * (it does not have the flag ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR).
  */
 HWLOC_DECLSPEC int
 hwloc_memattr_get_best_initiator(hwloc_topology_t topology,
                                  hwloc_memattr_id_t attribute,
-                                 hwloc_obj_t target,
+                                 hwloc_obj_t target_node,
                                  unsigned long flags,
                                  struct hwloc_location *best_initiator, hwloc_uint64_t *value);
 
-/** @} */
-
-
-/** \defgroup hwlocality_memattrs_manage Managing memory attributes
- * @{
- */
-
-/** \brief Return the name of a memory attribute.
- */
-HWLOC_DECLSPEC int
-hwloc_memattr_get_name(hwloc_topology_t topology,
-                       hwloc_memattr_id_t attribute,
-                       const char **name);
-
-/** \brief Return the flags of the given attribute.
- *
- * Flags are a OR'ed set of ::hwloc_memattr_flag_e.
- */
-HWLOC_DECLSPEC int
-hwloc_memattr_get_flags(hwloc_topology_t topology,
-                        hwloc_memattr_id_t attribute,
-                        unsigned long *flags);
-
-/** \brief Memory attribute flags.
- * Given to hwloc_memattr_register() and returned by hwloc_memattr_get_flags().
- */
-enum hwloc_memattr_flag_e {
-  /** \brief The best nodes for this memory attribute are those with the higher values.
-   * For instance Bandwidth.
-   */
-  HWLOC_MEMATTR_FLAG_HIGHER_FIRST = (1UL<<0),
-  /** \brief The best nodes for this memory attribute are those with the lower values.
-   * For instance Latency.
-   */
-  HWLOC_MEMATTR_FLAG_LOWER_FIRST = (1UL<<1),
-  /** \brief The value returned for this memory attribute depends on the given initiator.
-   * For instance Bandwidth and Latency, but not Capacity.
-   */
-  HWLOC_MEMATTR_FLAG_NEED_INITIATOR = (1UL<<2)
-};
-
-/** \brief Register a new memory attribute.
- *
- * Add a specific memory attribute that is not defined in ::hwloc_memattr_id_e.
- * Flags are a OR'ed set of ::hwloc_memattr_flag_e. It must contain at least
- * one of ::HWLOC_MEMATTR_FLAG_HIGHER_FIRST or ::HWLOC_MEMATTR_FLAG_LOWER_FIRST.
- */
-HWLOC_DECLSPEC int
-hwloc_memattr_register(hwloc_topology_t topology,
-                       const char *name,
-                       unsigned long flags,
-                       hwloc_memattr_id_t *id);
-
-/** \brief Set an attribute value for a specific target NUMA node.
- *
- * If the attribute does not relate to a specific initiator
- * (it does not have the flag ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR),
- * location \p initiator is ignored and may be \c NULL.
- *
- * The initiator will be copied into the topology,
- * the caller should free anything allocated to store the initiator,
- * for instance the cpuset.
- *
- * \p flags must be \c 0 for now.
- *
- * \note The initiator \p initiator should be of type ::HWLOC_LOCATION_TYPE_CPUSET
- * when refering to accesses performed by CPU cores.
- * ::HWLOC_LOCATION_TYPE_OBJECT is currently unused internally by hwloc,
- * but users may for instance use it to provide custom information about
- * host memory accesses performed by GPUs.
- */
-HWLOC_DECLSPEC int
-hwloc_memattr_set_value(hwloc_topology_t topology,
-                        hwloc_memattr_id_t attribute,
-                        hwloc_obj_t target_node,
-                        struct hwloc_location *initiator,
-                        unsigned long flags,
-                        hwloc_uint64_t value);
-
 /** \brief Return the target NUMA nodes that have some values for a given attribute.
  *
  * Return targets for the given attribute in the \p targets array
@@ -397,8 +482,10 @@ hwloc_memattr_set_value(hwloc_topology_t topology,
  * NUMA nodes with hwloc_get_local_numanode_objs() and then look at their attribute
  * values.
  *
+ * \return 0 on success or -1 on error.
+ *
  * \note The initiator \p initiator should be of type ::HWLOC_LOCATION_TYPE_CPUSET
- * when refering to accesses performed by CPU cores.
+ * when referring to accesses performed by CPU cores.
  * ::HWLOC_LOCATION_TYPE_OBJECT is currently unused internally by hwloc,
  * but users may for instance use it to provide custom information about
  * host memory accesses performed by GPUs.
@@ -408,7 +495,7 @@ hwloc_memattr_get_targets(hwloc_topology_t topology,
                           hwloc_memattr_id_t attribute,
                           struct hwloc_location *initiator,
                           unsigned long flags,
-                          unsigned *nrp, hwloc_obj_t *targets, hwloc_uint64_t *values);
+                          unsigned *nr, hwloc_obj_t *targets, hwloc_uint64_t *values);
 
 /** \brief Return the initiators that have values for a given attribute for a specific target NUMA node.
  *
@@ -428,12 +515,16 @@ hwloc_memattr_get_targets(hwloc_topology_t topology,
  * The returned initiators should not be modified or freed,
  * they belong to the topology.
  *
+ * \p target_node cannot be \c NULL.
+ *
  * \p flags must be \c 0 for now.
  *
  * If the attribute does not relate to a specific initiator
  * (it does not have the flag ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR),
  * no initiator is returned.
  *
+ * \return 0 on success or -1 on error.
+ *
  * \note This function is meant for tools and debugging (listing internal information)
  * rather than for application queries. Applications should rather select useful
  * NUMA nodes with hwloc_get_local_numanode_objs() and then look at their attribute
@@ -445,6 +536,131 @@ hwloc_memattr_get_initiators(hwloc_topology_t topology,
                              hwloc_obj_t target_node,
                              unsigned long flags,
                              unsigned *nr, struct hwloc_location *initiators, hwloc_uint64_t *values);
+
+/** @} */
+
+
+/** \defgroup hwlocality_memattrs_manage Managing memory attributes
+ *
+ * Memory attribues are identified by an ID (::hwloc_memattr_id_t)
+ * and a name. hwloc_memattr_get_name() and hwloc_memattr_get_by_name()
+ * convert between them (or return error if the attribute does not exist).
+ *
+ * The set of valid ::hwloc_memattr_id_t is a contigous set starting at \c 0.
+ * It first contains predefined attributes, as listed
+ * in ::hwloc_memattr_id_e (from \c 0 to \c HWLOC_MEMATTR_ID_MAX-1).
+ * Then custom attributes may be dynamically registered with
+ * hwloc_memattr_register(). They will get the following IDs
+ * (\c HWLOC_MEMATTR_ID_MAX for the first one, etc.).
+ *
+ * To iterate over all valid attributes
+ * (either predefined or dynamically registered custom ones),
+ * one may iterate over IDs starting from \c 0 until hwloc_memattr_get_name()
+ * or hwloc_memattr_get_flags() returns an error.
+ *
+ * The values for an existing attribute or for custom dynamically registered ones
+ * may be set or modified with hwloc_memattr_set_value().
+ *
+ * @{
+ */
+
+/** \brief Return the name of a memory attribute.
+ *
+ * The output pointer \p name cannot be \c NULL.
+ *
+ * \return 0 on success.
+ * \return -1 with errno set to \c EINVAL if the attribute does not exist.
+ */
+HWLOC_DECLSPEC int
+hwloc_memattr_get_name(hwloc_topology_t topology,
+                       hwloc_memattr_id_t attribute,
+                       const char **name);
+
+/** \brief Return the flags of the given attribute.
+ *
+ * Flags are a OR'ed set of ::hwloc_memattr_flag_e.
+ *
+ * The output pointer \p flags cannot be \c NULL.
+ *
+ * \return 0 on success.
+ * \return -1 with errno set to \c EINVAL if the attribute does not exist.
+ */
+HWLOC_DECLSPEC int
+hwloc_memattr_get_flags(hwloc_topology_t topology,
+                        hwloc_memattr_id_t attribute,
+                        unsigned long *flags);
+
+/** \brief Memory attribute flags.
+ * Given to hwloc_memattr_register() and returned by hwloc_memattr_get_flags().
+ */
+enum hwloc_memattr_flag_e {
+  /** \brief The best nodes for this memory attribute are those with the higher values.
+   * For instance Bandwidth.
+   */
+  HWLOC_MEMATTR_FLAG_HIGHER_FIRST = (1UL<<0),
+  /** \brief The best nodes for this memory attribute are those with the lower values.
+   * For instance Latency.
+   */
+  HWLOC_MEMATTR_FLAG_LOWER_FIRST = (1UL<<1),
+  /** \brief The value returned for this memory attribute depends on the given initiator.
+   * For instance Bandwidth and Latency, but not Capacity.
+   */
+  HWLOC_MEMATTR_FLAG_NEED_INITIATOR = (1UL<<2)
+};
+
+/** \brief Register a new memory attribute.
+ *
+ * Add a new custom memory attribute.
+ * Flags are a OR'ed set of ::hwloc_memattr_flag_e. It must contain one of
+ * ::HWLOC_MEMATTR_FLAG_HIGHER_FIRST or ::HWLOC_MEMATTR_FLAG_LOWER_FIRST but not both.
+ *
+ * The new attribute \p id is immediately after the last existing attribute ID
+ * (which is either the ID of the last registered attribute if any,
+ * or the ID of the last predefined attribute in ::hwloc_memattr_id_e).
+ *
+ * \return 0 on success.
+ * \return -1 with errno set to \c EINVAL if an invalid set of flags is given.
+ * \return -1 with errno set to \c EBUSY if another attribute already uses this name.
+ */
+HWLOC_DECLSPEC int
+hwloc_memattr_register(hwloc_topology_t topology,
+                       const char *name,
+                       unsigned long flags,
+                       hwloc_memattr_id_t *id);
+
+/** \brief Set an attribute value for a specific target NUMA node.
+ *
+ * If the attribute does not relate to a specific initiator
+ * (it does not have the flag ::HWLOC_MEMATTR_FLAG_NEED_INITIATOR),
+ * location \p initiator is ignored and may be \c NULL.
+ *
+ * The initiator will be copied into the topology,
+ * the caller should free anything allocated to store the initiator,
+ * for instance the cpuset.
+ *
+ * \p target_node cannot be \c NULL.
+ *
+ * \p attribute cannot be ::HWLOC_MEMATTR_FLAG_ID_CAPACITY or
+ * ::HWLOC_MEMATTR_FLAG_ID_LOCALITY.
+ *
+ * \p flags must be \c 0 for now.
+ *
+ * \note The initiator \p initiator should be of type ::HWLOC_LOCATION_TYPE_CPUSET
+ * when referring to accesses performed by CPU cores.
+ * ::HWLOC_LOCATION_TYPE_OBJECT is currently unused internally by hwloc,
+ * but users may for instance use it to provide custom information about
+ * host memory accesses performed by GPUs.
+ *
+ * \return 0 on success or -1 on error.
+ */
+HWLOC_DECLSPEC int
+hwloc_memattr_set_value(hwloc_topology_t topology,
+                        hwloc_memattr_id_t attribute,
+                        hwloc_obj_t target_node,
+                        struct hwloc_location *initiator,
+                        unsigned long flags,
+                        hwloc_uint64_t value);
+
 /** @} */
 
 #ifdef __cplusplus

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

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2012-2020 Inria.  All rights reserved.
+ * Copyright © 2012-2023 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -39,7 +39,7 @@ extern "C" {
 /** \brief Get the CPU set of processors that are physically
  * close to NVML device \p device.
  *
- * Return the CPU set describing the locality of the NVML device \p device.
+ * Store in \p set the CPU-set describing the locality of the NVML device \p device.
  *
  * Topology \p topology and device \p device must match the local machine.
  * I/O devices detection and the NVML component are not needed in the topology.
@@ -51,6 +51,9 @@ extern "C" {
  *
  * This function is currently only implemented in a meaningful way for
  * Linux; other systems will simply get a full cpuset.
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if device information could not be found.
  */
 static __hwloc_inline int
 hwloc_nvml_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
@@ -88,8 +91,8 @@ hwloc_nvml_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
 /** \brief Get the hwloc OS device object corresponding to the
  * NVML device whose index is \p idx.
  *
- * Return the OS device object describing the NVML device whose
+ * \return The hwloc OS device object describing the NVML device whose index is \p idx.
- * index is \p idx. Returns NULL if there is none.
+ * \return \c NULL if none could be found.
  *
  * The topology \p topology does not necessarily have to match the current
  * machine. For instance the topology may be an XML import of a remote host.
@@ -114,8 +117,8 @@ hwloc_nvml_get_device_osdev_by_index(hwloc_topology_t topology, unsigned idx)
 
 /** \brief Get the hwloc OS device object corresponding to NVML device \p device.
  *
- * Return the hwloc OS device object that describes the given
+ * \return The hwloc OS device object that describes the given NVML device \p device.
- * NVML device \p device. Return NULL if there is none.
+ * \return \c NULL if none could be found.
  *
  * Topology \p topology and device \p device must match the local machine.
  * I/O devices detection and the NVML component must be enabled in the topology.

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

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2012-2020 Inria.  All rights reserved.
+ * Copyright © 2012-2023 Inria.  All rights reserved.
  * Copyright © 2013, 2018 Université Bordeaux.  All right reserved.
  * See COPYING in top-level directory.
  */
@@ -41,6 +41,15 @@ extern "C" {
  */
 /* Copyright (c) 2008-2018 The Khronos Group Inc. */
 
+/* needs "cl_khr_pci_bus_info" device extension, but not strictly required for clGetDeviceInfo() */
+typedef struct {
+    cl_uint pci_domain;
+    cl_uint pci_bus;
+    cl_uint pci_device;
+    cl_uint pci_function;
+} hwloc_cl_device_pci_bus_info_khr;
+#define HWLOC_CL_DEVICE_PCI_BUS_INFO_KHR 0x410F
+
 /* needs "cl_amd_device_attribute_query" device extension, but not strictly required for clGetDeviceInfo() */
 #define HWLOC_CL_DEVICE_TOPOLOGY_AMD 0x4037
 typedef union {
@@ -69,22 +78,36 @@ typedef union {
 /** \brief Return the domain, bus and device IDs of the OpenCL device \p device.
  *
  * Device \p device must match the local machine.
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if device information could not be found.
  */
 static __hwloc_inline int
 hwloc_opencl_get_device_pci_busid(cl_device_id device,
                                unsigned *domain, unsigned *bus, unsigned *dev, unsigned *func)
 {
 	hwloc_cl_device_topology_amd amdtopo;
+	hwloc_cl_device_pci_bus_info_khr khrbusinfo;
 	cl_uint nvbus, nvslot, nvdomain;
 	cl_int clret;
 
+	clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_PCI_BUS_INFO_KHR, sizeof(khrbusinfo), &khrbusinfo, NULL);
+	if (CL_SUCCESS == clret) {
+		*domain = (unsigned) khrbusinfo.pci_domain;
+		*bus = (unsigned) khrbusinfo.pci_bus;
+		*dev = (unsigned) khrbusinfo.pci_device;
+		*func = (unsigned) khrbusinfo.pci_function;
+		return 0;
+	}
+
 	clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_TOPOLOGY_AMD, sizeof(amdtopo), &amdtopo, NULL);
 	if (CL_SUCCESS == clret
 	    && HWLOC_CL_DEVICE_TOPOLOGY_TYPE_PCIE_AMD == amdtopo.raw.type) {
 		*domain = 0; /* can't do anything better */
-		*bus = (unsigned) amdtopo.pcie.bus;
+		/* cl_device_topology_amd stores bus ID in cl_char, dont convert those signed char directly to unsigned int */
-		*dev = (unsigned) amdtopo.pcie.device;
+		*bus = (unsigned) (unsigned char) amdtopo.pcie.bus;
-		*func = (unsigned) amdtopo.pcie.function;
+		*dev = (unsigned) (unsigned char) amdtopo.pcie.device;
+		*func = (unsigned) (unsigned char) amdtopo.pcie.function;
 		return 0;
 	}
 
@@ -112,7 +135,7 @@ hwloc_opencl_get_device_pci_busid(cl_device_id device,
 /** \brief Get the CPU set of processors that are physically
  * close to OpenCL device \p device.
  *
- * Return the CPU set describing the locality of the OpenCL device \p device.
+ * Store in \p set the CPU-set describing the locality of the OpenCL device \p device.
  *
  * Topology \p topology and device \p device must match the local machine.
  * I/O devices detection and the OpenCL component are not needed in the topology.
@@ -125,6 +148,9 @@ hwloc_opencl_get_device_pci_busid(cl_device_id device,
  * This function is currently only implemented in a meaningful way for
  * Linux with the AMD or NVIDIA OpenCL implementation; other systems will simply
  * get a full cpuset.
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if the device could not be found.
  */
 static __hwloc_inline int
 hwloc_opencl_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
@@ -161,10 +187,10 @@ hwloc_opencl_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unuse
 /** \brief Get the hwloc OS device object corresponding to the
  * OpenCL device for the given indexes.
  *
- * Return the OS device object describing the OpenCL device
+ * \return The hwloc OS device object describing the OpenCL device
  * whose platform index is \p platform_index,
  * and whose device index within this platform if \p device_index.
- * Return NULL if there is none.
+ * \return \c NULL if there is none.
  *
  * The topology \p topology does not necessarily have to match the current
  * machine. For instance the topology may be an XML import of a remote host.
@@ -191,8 +217,9 @@ hwloc_opencl_get_device_osdev_by_index(hwloc_topology_t topology,
 
 /** \brief Get the hwloc OS device object corresponding to OpenCL device \p deviceX.
  *
- * Use OpenCL device attributes to find the corresponding hwloc OS device object.
+ * \return The hwloc OS device object corresponding to the given OpenCL device \p device.
- * Return NULL if there is none or if useful attributes are not available.
+ * \return \c NULL if none could be found, for instance
+ * if required OpenCL attributes are not available.
  *
  * This function currently only works on AMD and NVIDIA OpenCL devices that support
  * relevant OpenCL extensions. hwloc_opencl_get_device_osdev_by_index()

src/3rdparty/hwloc/include/hwloc/openfabrics-verbs.h

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2023 Inria.  All rights reserved.
  * Copyright © 2009-2010 Université Bordeaux
  * Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -44,7 +44,7 @@ extern "C" {
 /** \brief Get the CPU set of processors that are physically
  * close to device \p ibdev.
  *
- * Return the CPU set describing the locality of the OpenFabrics
+ * Store in \p set the CPU-set describing the locality of the OpenFabrics
  * device \p ibdev (InfiniBand, etc).
  *
  * Topology \p topology and device \p ibdev must match the local machine.
@@ -57,6 +57,9 @@ extern "C" {
  *
  * This function is currently only implemented in a meaningful way for
  * Linux; other systems will simply get a full cpuset.
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if device information could not be found.
  */
 static __hwloc_inline int
 hwloc_ibv_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
@@ -88,10 +91,11 @@ hwloc_ibv_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
 /** \brief Get the hwloc OS device object corresponding to the OpenFabrics
  * device named \p ibname.
  *
- * Return the OS device object describing the OpenFabrics device
+ * \return The hwloc OS device object describing the OpenFabrics device
  * (InfiniBand, Omni-Path, usNIC, etc) whose name is \p ibname
  * (mlx5_0, hfi1_0, usnic_0, qib0, etc).
- * Returns NULL if there is none.
+ * \return \c NULL if none could be found.
+ *
  * The name \p ibname is usually obtained from ibv_get_device_name().
  *
  * The topology \p topology does not necessarily have to match the current
@@ -117,8 +121,9 @@ hwloc_ibv_get_device_osdev_by_name(hwloc_topology_t topology,
 /** \brief Get the hwloc OS device object corresponding to the OpenFabrics
  * device \p ibdev.
  *
- * Return the OS device object describing the OpenFabrics device \p ibdev
+ * \return The hwloc OS device object describing the OpenFabrics
- * (InfiniBand, etc). Returns NULL if there is none.
+ * device \p ibdev (InfiniBand, etc).
+ * \return \c NULL if none could be found.
  *
  * Topology \p topology and device \p ibdev must match the local machine.
  * I/O devices detection must be enabled in the topology.

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

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2013-2020 Inria.  All rights reserved.
+ * Copyright © 2013-2024 Inria.  All rights reserved.
  * Copyright © 2016 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
  */
@@ -26,7 +26,10 @@ struct hwloc_backend;
 
 
 
-/** \defgroup hwlocality_disc_components Components and Plugins: Discovery components
+/** \defgroup hwlocality_disc_components Components and Plugins: Discovery components and backends
+ *
+ * \note These structures and functions may change when ::HWLOC_COMPONENT_ABI is modified.
+ *
  * @{
  */
 
@@ -87,15 +90,6 @@ struct hwloc_disc_component {
   struct hwloc_disc_component * next;
 };
 
-/** @} */
-
-
-
-
-/** \defgroup hwlocality_disc_backends Components and Plugins: Discovery backends
- * @{
- */
-
 /** \brief Discovery phase */
 typedef enum hwloc_disc_phase_e {
   /** \brief xml or synthetic, platform-specific components such as bgq.
@@ -158,7 +152,7 @@ struct hwloc_disc_status {
    */
   unsigned excluded_phases;
 
-  /** \brief OR'ed set of hwloc_disc_status_flag_e */
+  /** \brief OR'ed set of ::hwloc_disc_status_flag_e */
   unsigned long flags;
 };
 
@@ -241,6 +235,9 @@ HWLOC_DECLSPEC int hwloc_backend_enable(struct hwloc_backend *backend);
 
 
 /** \defgroup hwlocality_generic_components Components and Plugins: Generic components
+ *
+ * \note These structures and functions may change when ::HWLOC_COMPONENT_ABI is modified.
+ *
  * @{
  */
 
@@ -304,18 +301,98 @@ struct hwloc_component {
   void * data;
 };
 
+/** \brief Make sure that plugins can lookup core symbols.
+ *
+ * This is a sanity check to avoid lazy-lookup failures when libhwloc
+ * is loaded within a plugin, and later tries to load its own plugins.
+ * This may fail (and abort the program) if libhwloc symbols are in a
+ * private namespace.
+ *
+ * \return 0 on success.
+ * \return -1 if the plugin cannot be successfully loaded. The caller
+ * plugin init() callback should return a negative error code as well.
+ *
+ * Plugins should call this function in their init() callback to avoid
+ * later crashes if lazy symbol resolution is used by the upper layer that
+ * loaded hwloc (e.g. OpenCL implementations using dlopen with RTLD_LAZY).
+ *
+ * \note The build system must define HWLOC_INSIDE_PLUGIN if and only if
+ * building the caller as a plugin.
+ *
+ * \note This function should remain inline so plugins can call it even
+ * when they cannot find libhwloc symbols.
+ */
+static __hwloc_inline int
+hwloc_plugin_check_namespace(const char *pluginname __hwloc_attribute_unused, const char *symbol __hwloc_attribute_unused)
+{
+#ifdef HWLOC_INSIDE_PLUGIN
+  void *sym;
+#ifdef HWLOC_HAVE_LTDL
+  lt_dlhandle handle = lt_dlopen(NULL);
+#else
+  void *handle = dlopen(NULL, RTLD_NOW|RTLD_LOCAL);
+#endif
+  if (!handle)
+    /* cannot check, assume things will work */
+    return 0;
+#ifdef HWLOC_HAVE_LTDL
+  sym = lt_dlsym(handle, symbol);
+  lt_dlclose(handle);
+#else
+  sym = dlsym(handle, symbol);
+  dlclose(handle);
+#endif
+  if (!sym) {
+    static int verboseenv_checked = 0;
+    static int verboseenv_value = 0;
+    if (!verboseenv_checked) {
+      const char *verboseenv = getenv("HWLOC_PLUGINS_VERBOSE");
+      verboseenv_value = verboseenv ? atoi(verboseenv) : 0;
+      verboseenv_checked = 1;
+    }
+    if (verboseenv_value)
+      fprintf(stderr, "Plugin `%s' disabling itself because it cannot find the `%s' core symbol.\n",
+	      pluginname, symbol);
+    return -1;
+  }
+#endif /* HWLOC_INSIDE_PLUGIN */
+  return 0;
+}
+
 /** @} */
 
 
 
 
 /** \defgroup hwlocality_components_core_funcs Components and Plugins: Core functions to be used by components
+ *
+ * \note These structures and functions may change when ::HWLOC_COMPONENT_ABI is modified.
+ *
  * @{
  */
 
-/** \brief Check whether insertion errors are hidden */
+/** \brief Check whether error messages are hidden.
+ *
+ * Callers should print critical error messages
+ * (e.g. invalid hw topo info, invalid config)
+ * only if this function returns strictly less than 2.
+ *
+ * Callers should print non-critical error messages
+ * (e.g. failure to initialize CUDA)
+ * if this function returns 0.
+ *
+ * This function return 1 by default (show critical only),
+ * 0 in lstopo (show all),
+ * or anything set in HWLOC_HIDE_ERRORS in the environment.
+ *
+ * Use macros HWLOC_SHOW_CRITICAL_ERRORS() and HWLOC_SHOW_ALL_ERRORS()
+ * for clarity.
+ */
 HWLOC_DECLSPEC int hwloc_hide_errors(void);
 
+#define HWLOC_SHOW_CRITICAL_ERRORS() (hwloc_hide_errors() < 2)
+#define HWLOC_SHOW_ALL_ERRORS() (hwloc_hide_errors() == 0)
+
 /** \brief Add an object to the topology.
  *
  * Insert new object \p obj in the topology starting under existing object \p root
@@ -391,70 +468,15 @@ HWLOC_DECLSPEC int hwloc_obj_add_children_sets(hwloc_obj_t obj);
  */
 HWLOC_DECLSPEC int hwloc_topology_reconnect(hwloc_topology_t topology, unsigned long flags __hwloc_attribute_unused);
 
-/** \brief Make sure that plugins can lookup core symbols.
- *
- * This is a sanity check to avoid lazy-lookup failures when libhwloc
- * is loaded within a plugin, and later tries to load its own plugins.
- * This may fail (and abort the program) if libhwloc symbols are in a
- * private namespace.
- *
- * \return 0 on success.
- * \return -1 if the plugin cannot be successfully loaded. The caller
- * plugin init() callback should return a negative error code as well.
- *
- * Plugins should call this function in their init() callback to avoid
- * later crashes if lazy symbol resolution is used by the upper layer that
- * loaded hwloc (e.g. OpenCL implementations using dlopen with RTLD_LAZY).
- *
- * \note The build system must define HWLOC_INSIDE_PLUGIN if and only if
- * building the caller as a plugin.
- *
- * \note This function should remain inline so plugins can call it even
- * when they cannot find libhwloc symbols.
- */
-static __hwloc_inline int
-hwloc_plugin_check_namespace(const char *pluginname __hwloc_attribute_unused, const char *symbol __hwloc_attribute_unused)
-{
-#ifdef HWLOC_INSIDE_PLUGIN
-  void *sym;
-#ifdef HWLOC_HAVE_LTDL
-  lt_dlhandle handle = lt_dlopen(NULL);
-#else
-  void *handle = dlopen(NULL, RTLD_NOW|RTLD_LOCAL);
-#endif
-  if (!handle)
-    /* cannot check, assume things will work */
-    return 0;
-#ifdef HWLOC_HAVE_LTDL
-  sym = lt_dlsym(handle, symbol);
-  lt_dlclose(handle);
-#else
-  sym = dlsym(handle, symbol);
-  dlclose(handle);
-#endif
-  if (!sym) {
-    static int verboseenv_checked = 0;
-    static int verboseenv_value = 0;
-    if (!verboseenv_checked) {
-      const char *verboseenv = getenv("HWLOC_PLUGINS_VERBOSE");
-      verboseenv_value = verboseenv ? atoi(verboseenv) : 0;
-      verboseenv_checked = 1;
-    }
-    if (verboseenv_value)
-      fprintf(stderr, "Plugin `%s' disabling itself because it cannot find the `%s' core symbol.\n",
-	      pluginname, symbol);
-    return -1;
-  }
-#endif /* HWLOC_INSIDE_PLUGIN */
-  return 0;
-}
-
 /** @} */
 
 
 
 
 /** \defgroup hwlocality_components_filtering Components and Plugins: Filtering objects
+ *
+ * \note These structures and functions may change when ::HWLOC_COMPONENT_ABI is modified.
+ *
  * @{
  */
 
@@ -469,9 +491,12 @@ hwloc_filter_check_pcidev_subtype_important(unsigned classid)
   return (baseclass == 0x03 /* PCI_BASE_CLASS_DISPLAY */
 	  || baseclass == 0x02 /* PCI_BASE_CLASS_NETWORK */
 	  || baseclass == 0x01 /* PCI_BASE_CLASS_STORAGE */
+	  || baseclass == 0x00 /* Unclassified, for Atos/Bull BXI */
 	  || baseclass == 0x0b /* PCI_BASE_CLASS_PROCESSOR */
 	  || classid == 0x0c04 /* PCI_CLASS_SERIAL_FIBER */
 	  || classid == 0x0c06 /* PCI_CLASS_SERIAL_INFINIBAND */
+          || classid == 0x0502 /* PCI_CLASS_MEMORY_CXL */
+          || baseclass == 0x06 /* PCI_BASE_CLASS_BRIDGE with non-PCI downstream. the core will drop the useless ones later */
 	  || baseclass == 0x12 /* Processing Accelerators */);
 }
 
@@ -527,6 +552,9 @@ hwloc_filter_check_keep_object(hwloc_topology_t topology, hwloc_obj_t obj)
 
 
 /** \defgroup hwlocality_components_pcidisc Components and Plugins: helpers for PCI discovery
+ *
+ * \note These structures and functions may change when ::HWLOC_COMPONENT_ABI is modified.
+ *
  * @{
  */
 
@@ -578,18 +606,89 @@ HWLOC_DECLSPEC int hwloc_pcidisc_tree_attach(struct hwloc_topology *topology, st
 
 
 /** \defgroup hwlocality_components_pcifind Components and Plugins: finding PCI objects during other discoveries
+ *
+ * \note These structures and functions may change when ::HWLOC_COMPONENT_ABI is modified.
+ *
  * @{
  */
 
-/** \brief Find the normal parent of a PCI bus ID.
+/** \brief Find the object or a parent of a PCI bus ID.
  *
- * Look at PCI affinity to find out where the given PCI bus ID should be attached.
+ * When attaching a new object (typically an OS device) whose locality
+ * is specified by PCI bus ID, this function returns the PCI object
+ * to use as a parent for attaching.
  *
- * This function should be used to attach an I/O device under the corresponding
+ * If the exact PCI device with this bus ID exists, it is returned.
- * PCI object (if any), or under a normal (non-I/O) object with same locality.
+ * Otherwise (for instance if it was filtered out), the function returns
+ * another object with similar locality (for instance a parent bridge,
+ * or the local CPU Package).
  */
 HWLOC_DECLSPEC struct hwloc_obj * hwloc_pci_find_parent_by_busid(struct hwloc_topology *topology, unsigned domain, unsigned bus, unsigned dev, unsigned func);
 
+/** \brief Find the PCI device or bridge matching a PCI bus ID exactly.
+ *
+ * This is useful for adding specific information about some objects
+ * based on their PCI id. When it comes to attaching objects based on
+ * PCI locality, hwloc_pci_find_parent_by_busid() should be preferred.
+ */
+HWLOC_DECLSPEC struct hwloc_obj * hwloc_pci_find_by_busid(struct hwloc_topology *topology, unsigned domain, unsigned bus, unsigned dev, unsigned func);
+
+
+/** @} */
+
+
+
+
+/** \defgroup hwlocality_components_distances Components and Plugins: distances
+ *
+ * \note These structures and functions may change when ::HWLOC_COMPONENT_ABI is modified.
+ *
+ * @{
+ */
+
+/** \brief Handle to a new distances structure during its addition to the topology. */
+typedef void * hwloc_backend_distances_add_handle_t;
+
+/** \brief Create a new empty distances structure.
+ *
+ * This is identical to hwloc_distances_add_create()
+ * but this variant is designed for backend inserting
+ * distances during topology discovery.
+ */
+HWLOC_DECLSPEC hwloc_backend_distances_add_handle_t
+hwloc_backend_distances_add_create(hwloc_topology_t topology,
+                                   const char *name, unsigned long kind,
+                                   unsigned long flags);
+
+/** \brief Specify the objects and values in a new empty distances structure.
+ *
+ * This is similar to hwloc_distances_add_values()
+ * but this variant is designed for backend inserting
+ * distances during topology discovery.
+ *
+ * The only semantical difference is that \p objs and \p values
+ * are not duplicated, but directly attached to the topology.
+ * On success, these arrays are given to the core and should not
+ * ever be freed by the caller anymore.
+ */
+HWLOC_DECLSPEC int
+hwloc_backend_distances_add_values(hwloc_topology_t topology,
+                                   hwloc_backend_distances_add_handle_t handle,
+                                   unsigned nbobjs, hwloc_obj_t *objs,
+                                   hwloc_uint64_t *values,
+                                   unsigned long flags);
+
+/** \brief Commit a new distances structure.
+ *
+ * This is similar to hwloc_distances_add_commit()
+ * but this variant is designed for backend inserting
+ * distances during topology discovery.
+ */
+HWLOC_DECLSPEC int
+hwloc_backend_distances_add_commit(hwloc_topology_t topology,
+                                   hwloc_backend_distances_add_handle_t handle,
+                                   unsigned long flags);
+
 /** @} */
 
 

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
- * Copyright © 2010-2020 Inria.  All rights reserved.
+ * Copyright © 2010-2025 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -120,6 +120,12 @@ extern "C" {
 #define HWLOC_TOPOLOGY_FLAG_IS_THISSYSTEM HWLOC_NAME_CAPS(TOPOLOGY_FLAG_IS_THISSYSTEM)
 #define HWLOC_TOPOLOGY_FLAG_THISSYSTEM_ALLOWED_RESOURCES HWLOC_NAME_CAPS(TOPOLOGY_FLAG_THISSYSTEM_ALLOWED_RESOURCES)
 #define HWLOC_TOPOLOGY_FLAG_IMPORT_SUPPORT HWLOC_NAME_CAPS(TOPOLOGY_FLAG_IMPORT_SUPPORT)
+#define HWLOC_TOPOLOGY_FLAG_RESTRICT_TO_CPUBINDING HWLOC_NAME_CAPS(TOPOLOGY_FLAG_RESTRICT_TO_CPUBINDING)
+#define HWLOC_TOPOLOGY_FLAG_RESTRICT_TO_MEMBINDING HWLOC_NAME_CAPS(TOPOLOGY_FLAG_RESTRICT_TO_MEMBINDING)
+#define HWLOC_TOPOLOGY_FLAG_DONT_CHANGE_BINDING HWLOC_NAME_CAPS(TOPOLOGY_FLAG_DONT_CHANGE_BINDING)
+#define HWLOC_TOPOLOGY_FLAG_NO_DISTANCES HWLOC_NAME_CAPS(TOPOLOGY_FLAG_NO_DISTANCES)
+#define HWLOC_TOPOLOGY_FLAG_NO_MEMATTRS HWLOC_NAME_CAPS(TOPOLOGY_FLAG_NO_MEMATTRS)
+#define HWLOC_TOPOLOGY_FLAG_NO_CPUKINDS HWLOC_NAME_CAPS(TOPOLOGY_FLAG_NO_CPUKINDS)
 
 #define hwloc_topology_set_pid HWLOC_NAME(topology_set_pid)
 #define hwloc_topology_set_synthetic HWLOC_NAME(topology_set_synthetic)
@@ -170,6 +176,7 @@ extern "C" {
 
 #define hwloc_topology_insert_misc_object HWLOC_NAME(topology_insert_misc_object)
 #define hwloc_topology_alloc_group_object HWLOC_NAME(topology_alloc_group_object)
+#define hwloc_topology_free_group_object HWLOC_NAME(topology_free_group_object)
 #define hwloc_topology_insert_group_object HWLOC_NAME(topology_insert_group_object)
 #define hwloc_obj_add_other_obj_sets HWLOC_NAME(obj_add_other_obj_sets)
 #define hwloc_topology_refresh HWLOC_NAME(topology_refresh)
@@ -203,6 +210,7 @@ extern "C" {
 
 #define hwloc_obj_get_info_by_name HWLOC_NAME(obj_get_info_by_name)
 #define hwloc_obj_add_info HWLOC_NAME(obj_add_info)
+#define hwloc_obj_set_subtype HWLOC_NAME(obj_set_subtype)
 
 #define HWLOC_CPUBIND_PROCESS HWLOC_NAME_CAPS(CPUBIND_PROCESS)
 #define HWLOC_CPUBIND_THREAD HWLOC_NAME_CAPS(CPUBIND_THREAD)
@@ -225,6 +233,7 @@ extern "C" {
 #define HWLOC_MEMBIND_FIRSTTOUCH HWLOC_NAME_CAPS(MEMBIND_FIRSTTOUCH)
 #define HWLOC_MEMBIND_BIND HWLOC_NAME_CAPS(MEMBIND_BIND)
 #define HWLOC_MEMBIND_INTERLEAVE HWLOC_NAME_CAPS(MEMBIND_INTERLEAVE)
+#define HWLOC_MEMBIND_WEIGHTED_INTERLEAVE HWLOC_NAME_CAPS(MEMBIND_WEIGHTED_INTERLEAVE)
 #define HWLOC_MEMBIND_NEXTTOUCH HWLOC_NAME_CAPS(MEMBIND_NEXTTOUCH)
 #define HWLOC_MEMBIND_MIXED HWLOC_NAME_CAPS(MEMBIND_MIXED)
 
@@ -356,6 +365,7 @@ extern "C" {
 #define hwloc_get_closest_objs HWLOC_NAME(get_closest_objs)
 #define hwloc_get_obj_below_by_type HWLOC_NAME(get_obj_below_by_type)
 #define hwloc_get_obj_below_array_by_type HWLOC_NAME(get_obj_below_array_by_type)
+#define hwloc_get_obj_with_same_locality HWLOC_NAME(get_obj_with_same_locality)
 #define hwloc_distrib_flags_e HWLOC_NAME(distrib_flags_e)
 #define HWLOC_DISTRIB_FLAG_REVERSE HWLOC_NAME_CAPS(DISTRIB_FLAG_REVERSE)
 #define hwloc_distrib HWLOC_NAME(distrib)
@@ -377,6 +387,11 @@ extern "C" {
 #define HWLOC_MEMATTR_ID_LOCALITY HWLOC_NAME_CAPS(MEMATTR_ID_LOCALITY)
 #define HWLOC_MEMATTR_ID_BANDWIDTH HWLOC_NAME_CAPS(MEMATTR_ID_BANDWIDTH)
 #define HWLOC_MEMATTR_ID_LATENCY HWLOC_NAME_CAPS(MEMATTR_ID_LATENCY)
+#define HWLOC_MEMATTR_ID_READ_BANDWIDTH HWLOC_NAME_CAPS(MEMATTR_ID_READ_BANDWIDTH)
+#define HWLOC_MEMATTR_ID_WRITE_BANDWIDTH HWLOC_NAME_CAPS(MEMATTR_ID_WRITE_BANDWIDTH)
+#define HWLOC_MEMATTR_ID_READ_LATENCY HWLOC_NAME_CAPS(MEMATTR_ID_READ_LATENCY)
+#define HWLOC_MEMATTR_ID_WRITE_LATENCY HWLOC_NAME_CAPS(MEMATTR_ID_WRITE_LATENCY)
+#define HWLOC_MEMATTR_ID_MAX HWLOC_NAME_CAPS(MEMATTR_ID_MAX)
 
 #define hwloc_memattr_id_t HWLOC_NAME(memattr_id_t)
 #define hwloc_memattr_get_by_name HWLOC_NAME(memattr_get_by_name)
@@ -394,8 +409,10 @@ extern "C" {
 #define hwloc_local_numanode_flag_e HWLOC_NAME(local_numanode_flag_e)
 #define HWLOC_LOCAL_NUMANODE_FLAG_LARGER_LOCALITY HWLOC_NAME_CAPS(LOCAL_NUMANODE_FLAG_LARGER_LOCALITY)
 #define HWLOC_LOCAL_NUMANODE_FLAG_SMALLER_LOCALITY HWLOC_NAME_CAPS(LOCAL_NUMANODE_FLAG_SMALLER_LOCALITY)
+#define HWLOC_LOCAL_NUMANODE_FLAG_INTERSECT_LOCALITY HWLOC_NAME_CAPS(LOCAL_NUMANODE_FLAG_INTERSECT_LOCALITY)
 #define HWLOC_LOCAL_NUMANODE_FLAG_ALL HWLOC_NAME_CAPS(LOCAL_NUMANODE_FLAG_ALL)
 #define hwloc_get_local_numanode_objs HWLOC_NAME(get_local_numanode_objs)
+#define hwloc_topology_get_default_nodeset HWLOC_NAME(topology_get_default_nodeset)
 
 #define hwloc_memattr_get_name HWLOC_NAME(memattr_get_name)
 #define hwloc_memattr_get_flags HWLOC_NAME(memattr_get_flags)
@@ -454,11 +471,22 @@ extern "C" {
 #define hwloc_distances_obj_index HWLOC_NAME(distances_obj_index)
 #define hwloc_distances_obj_pair_values HWLOC_NAME(distances_pair_values)
 
+#define hwloc_distances_transform_e HWLOC_NAME(distances_transform_e)
+#define HWLOC_DISTANCES_TRANSFORM_REMOVE_NULL HWLOC_NAME_CAPS(DISTANCES_TRANSFORM_REMOVE_NULL)
+#define HWLOC_DISTANCES_TRANSFORM_LINKS HWLOC_NAME_CAPS(DISTANCES_TRANSFORM_LINKS)
+#define HWLOC_DISTANCES_TRANSFORM_MERGE_SWITCH_PORTS HWLOC_NAME_CAPS(DISTANCES_TRANSFORM_MERGE_SWITCH_PORTS)
+#define HWLOC_DISTANCES_TRANSFORM_TRANSITIVE_CLOSURE HWLOC_NAME_CAPS(DISTANCES_TRANSFORM_TRANSITIVE_CLOSURE)
+#define hwloc_distances_transform HWLOC_NAME(distances_transform)
+
 #define hwloc_distances_add_flag_e HWLOC_NAME(distances_add_flag_e)
 #define HWLOC_DISTANCES_ADD_FLAG_GROUP HWLOC_NAME_CAPS(DISTANCES_ADD_FLAG_GROUP)
 #define HWLOC_DISTANCES_ADD_FLAG_GROUP_INACCURATE HWLOC_NAME_CAPS(DISTANCES_ADD_FLAG_GROUP_INACCURATE)
 
-#define hwloc_distances_add HWLOC_NAME(distances_add)
+#define hwloc_distances_add_handle_t HWLOC_NAME(distances_add_handle_t)
+#define hwloc_distances_add_create HWLOC_NAME(distances_add_create)
+#define hwloc_distances_add_values HWLOC_NAME(distances_add_values)
+#define hwloc_distances_add_commit HWLOC_NAME(distances_add_commit)
+
 #define hwloc_distances_remove HWLOC_NAME(distances_remove)
 #define hwloc_distances_remove_by_depth HWLOC_NAME(distances_remove_by_depth)
 #define hwloc_distances_remove_by_type HWLOC_NAME(distances_remove_by_type)
@@ -523,6 +551,11 @@ extern "C" {
 #define hwloc_linux_get_tid_last_cpu_location HWLOC_NAME(linux_get_tid_last_cpu_location)
 #define hwloc_linux_read_path_as_cpumask HWLOC_NAME(linux_read_file_cpumask)
 
+/* windows.h */
+
+#define hwloc_windows_get_nr_processor_groups HWLOC_NAME(windows_get_nr_processor_groups)
+#define hwloc_windows_get_processor_group_cpuset HWLOC_NAME(windows_get_processor_group_cpuset)
+
 /* openfabrics-verbs.h */
 
 #define hwloc_ibv_get_device_cpuset HWLOC_NAME(ibv_get_device_cpuset)
@@ -531,6 +564,7 @@ extern "C" {
 
 /* opencl.h */
 
+#define hwloc_cl_device_pci_bus_info_khr HWLOC_NAME(cl_device_pci_bus_info_khr)
 #define hwloc_cl_device_topology_amd HWLOC_NAME(cl_device_topology_amd)
 #define hwloc_opencl_get_device_pci_busid HWLOC_NAME(opencl_get_device_pci_ids)
 #define hwloc_opencl_get_device_cpuset HWLOC_NAME(opencl_get_device_cpuset)
@@ -564,6 +598,13 @@ extern "C" {
 #define hwloc_rsmi_get_device_osdev HWLOC_NAME(rsmi_get_device_osdev)
 #define hwloc_rsmi_get_device_osdev_by_index HWLOC_NAME(rsmi_get_device_osdev_by_index)
 
+/* levelzero.h */
+
+#define hwloc_levelzero_get_device_cpuset HWLOC_NAME(levelzero_get_device_cpuset)
+#define hwloc_levelzero_get_sysman_device_cpuset HWLOC_NAME(levelzero_get_sysman_device_cpuset)
+#define hwloc_levelzero_get_device_osdev HWLOC_NAME(levelzero_get_device_osdev)
+#define hwloc_levelzero_get_sysman_device_osdev HWLOC_NAME(levelzero_get_sysman_device_osdev)
+
 /* gl.h */
 
 #define hwloc_gl_get_display_osdev_by_port_device HWLOC_NAME(gl_get_display_osdev_by_port_device)
@@ -620,10 +661,18 @@ extern "C" {
 #define hwloc_pcidisc_tree_insert_by_busid HWLOC_NAME(pcidisc_tree_insert_by_busid)
 #define hwloc_pcidisc_tree_attach HWLOC_NAME(pcidisc_tree_attach)
 
+#define hwloc_pci_find_by_busid HWLOC_NAME(pcidisc_find_by_busid)
 #define hwloc_pci_find_parent_by_busid HWLOC_NAME(pcidisc_find_busid_parent)
 
+#define hwloc_backend_distances_add_handle_t HWLOC_NAME(backend_distances_add_handle_t)
+#define hwloc_backend_distances_add_create HWLOC_NAME(backend_distances_add_create)
+#define hwloc_backend_distances_add_values HWLOC_NAME(backend_distances_add_values)
+#define hwloc_backend_distances_add_commit HWLOC_NAME(backend_distances_add_commit)
+
 /* hwloc/deprecated.h */
 
+#define hwloc_distances_add HWLOC_NAME(distances_add)
+
 #define hwloc_topology_insert_misc_object_by_parent HWLOC_NAME(topology_insert_misc_object_by_parent)
 #define hwloc_obj_cpuset_snprintf HWLOC_NAME(obj_cpuset_snprintf)
 #define hwloc_obj_type_sscanf HWLOC_NAME(obj_type_sscanf)
@@ -673,6 +722,8 @@ extern "C" {
 #define hwloc__obj_type_is_dcache HWLOC_NAME(_obj_type_is_dcache)
 #define hwloc__obj_type_is_icache HWLOC_NAME(_obj_type_is_icache)
 
+#define hwloc__pci_link_speed HWLOC_NAME(_pci_link_speed)
+
 /* private/cpuid-x86.h */
 
 #define hwloc_have_x86_cpuid HWLOC_NAME(have_x86_cpuid)
@@ -733,6 +784,7 @@ extern "C" {
 
 #define hwloc_cuda_component HWLOC_NAME(cuda_component)
 #define hwloc_gl_component HWLOC_NAME(gl_component)
+#define hwloc_levelzero_component HWLOC_NAME(levelzero_component)
 #define hwloc_nvml_component HWLOC_NAME(nvml_component)
 #define hwloc_rsmi_component HWLOC_NAME(rsmi_component)
 #define hwloc_opencl_component HWLOC_NAME(opencl_component)
@@ -765,6 +817,8 @@ extern "C" {
 #define hwloc_topology_setup_defaults HWLOC_NAME(topology_setup_defaults)
 #define hwloc_topology_clear HWLOC_NAME(topology_clear)
 
+#define hwloc__reconnect HWLOC_NAME(_reconnect)
+
 #define hwloc__attach_memory_object HWLOC_NAME(insert_memory_object)
 
 #define hwloc_get_obj_by_type_and_gp_index HWLOC_NAME(get_obj_by_type_and_gp_index)
@@ -772,7 +826,6 @@ extern "C" {
 #define hwloc_pci_discovery_init HWLOC_NAME(pci_discovery_init)
 #define hwloc_pci_discovery_prepare HWLOC_NAME(pci_discovery_prepare)
 #define hwloc_pci_discovery_exit HWLOC_NAME(pci_discovery_exit)
-#define hwloc_pci_find_by_busid HWLOC_NAME(pcidisc_find_by_busid)
 #define hwloc_find_insert_io_parent_by_complete_cpuset HWLOC_NAME(hwloc_find_insert_io_parent_by_complete_cpuset)
 
 #define hwloc__add_info HWLOC_NAME(_add_info)
@@ -816,7 +869,6 @@ extern "C" {
 #define hwloc_internal_distances_dup HWLOC_NAME(internal_distances_dup)
 #define hwloc_internal_distances_refresh HWLOC_NAME(internal_distances_refresh)
 #define hwloc_internal_distances_destroy HWLOC_NAME(internal_distances_destroy)
-
 #define hwloc_internal_distances_add HWLOC_NAME(internal_distances_add)
 #define hwloc_internal_distances_add_by_index HWLOC_NAME(internal_distances_add_by_index)
 #define hwloc_internal_distances_invalidate_cached_objs HWLOC_NAME(hwloc_internal_distances_invalidate_cached_objs)
@@ -830,6 +882,7 @@ extern "C" {
 #define hwloc_internal_memattrs_destroy HWLOC_NAME(internal_memattrs_destroy)
 #define hwloc_internal_memattrs_need_refresh HWLOC_NAME(internal_memattrs_need_refresh)
 #define hwloc_internal_memattrs_refresh HWLOC_NAME(internal_memattrs_refresh)
+#define hwloc_internal_memattrs_guess_memory_tiers HWLOC_NAME(internal_memattrs_guess_memory_tiers)
 
 #define hwloc_internal_cpukind_s HWLOC_NAME(internal_cpukind_s)
 #define hwloc_internal_cpukinds_init HWLOC_NAME(internal_cpukinds_init)

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

@@ -0,0 +1,206 @@
+/*
+ * Copyright © 2012-2023 Inria.  All rights reserved.
+ * Copyright (c) 2020, Advanced Micro Devices, Inc. All rights reserved.
+ * Written by Advanced Micro Devices,
+ * See COPYING in top-level directory.
+ */
+
+/** \file
+ * \brief Macros to help interaction between hwloc and the ROCm SMI Management Library.
+ *
+ * Applications that use both hwloc and the ROCm SMI Management Library may want to
+ * include this file so as to get topology information for AMD GPU devices.
+ */
+
+#ifndef HWLOC_RSMI_H
+#define HWLOC_RSMI_H
+
+#include "hwloc.h"
+#include "hwloc/autogen/config.h"
+#include "hwloc/helper.h"
+#ifdef HWLOC_LINUX_SYS
+#include "hwloc/linux.h"
+#endif
+
+#include <rocm_smi/rocm_smi.h>
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+/** \defgroup hwlocality_rsmi Interoperability with the ROCm SMI Management Library
+ *
+ * This interface offers ways to retrieve topology information about
+ * devices managed by the ROCm SMI Management Library.
+ *
+ * @{
+ */
+
+/** \brief Get the CPU set of logical processors that are physically
+ * close to AMD GPU device whose index is \p dv_ind.
+ *
+ * Store in \p set the CPU-set describing the locality of the AMD GPU device
+ * whose index is \p dv_ind.
+ *
+ * Topology \p topology and device \p dv_ind must match the local machine.
+ * I/O devices detection and the ROCm SMI component are not needed in the
+ * topology.
+ *
+ * The function only returns the locality of the device.
+ * If more information about the device is needed, OS objects should
+ * be used instead, see hwloc_rsmi_get_device_osdev()
+ * and hwloc_rsmi_get_device_osdev_by_index().
+ *
+ * This function is currently only implemented in a meaningful way for
+ * Linux; other systems will simply get a full cpuset.
+ *
+ * \return 0 on success.
+ * \return -1 on error, for instance if device information could not be found.
+ */
+static __hwloc_inline int
+hwloc_rsmi_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
+                             uint32_t dv_ind, hwloc_cpuset_t set)
+{
+#ifdef HWLOC_LINUX_SYS
+  /* If we're on Linux, use the sysfs mechanism to get the local cpus */
+#define HWLOC_RSMI_DEVICE_SYSFS_PATH_MAX 128
+  char path[HWLOC_RSMI_DEVICE_SYSFS_PATH_MAX];
+  rsmi_status_t ret;
+  uint64_t bdfid = 0;
+  unsigned domain, device, bus;
+
+  if (!hwloc_topology_is_thissystem(topology)) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  ret = rsmi_dev_pci_id_get(dv_ind, &bdfid);
+  if (RSMI_STATUS_SUCCESS != ret) {
+    errno = EINVAL;
+    return -1;
+  }
+  domain = (bdfid>>32) & 0xffffffff;
+  bus = ((bdfid & 0xffff)>>8) & 0xff;
+  device = ((bdfid & 0xff)>>3) & 0x1f;
+
+  sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.0/local_cpus", domain, bus, device);
+  if (hwloc_linux_read_path_as_cpumask(path, set) < 0
+      || hwloc_bitmap_iszero(set))
+    hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
+#else
+  /* Non-Linux systems simply get a full cpuset */
+  hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
+#endif
+  return 0;
+}
+
+/** \brief Get the hwloc OS device object corresponding to the
+ * AMD GPU device whose index is \p dv_ind.
+ *
+ * \return The hwloc OS device object describing the AMD GPU device whose
+ * index is \p dv_ind.
+ * \return \c NULL if none could be found.
+ *
+ * The topology \p topology does not necessarily have to match the current
+ * machine. For instance the topology may be an XML import of a remote host.
+ * I/O devices detection and the ROCm SMI component must be enabled in the
+ * topology.
+ *
+ * \note The corresponding PCI device object can be obtained by looking
+ * at the OS device parent object (unless PCI devices are filtered out).
+ */
+static __hwloc_inline hwloc_obj_t
+hwloc_rsmi_get_device_osdev_by_index(hwloc_topology_t topology, uint32_t dv_ind)
+{
+  hwloc_obj_t osdev = NULL;
+  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
+    if (HWLOC_OBJ_OSDEV_GPU == osdev->attr->osdev.type
+      && osdev->name
+      && !strncmp("rsmi", osdev->name, 4)
+      && atoi(osdev->name + 4) == (int) dv_ind)
+      return osdev;
+  }
+  return NULL;
+}
+
+/** \brief Get the hwloc OS device object corresponding to AMD GPU device,
+ * whose index is \p dv_ind.
+ *
+ * \return The hwloc OS device object that describes the given
+ * AMD GPU, whose index is \p dv_ind.
+ * \return \c NULL if none could be found.
+ *
+ * Topology \p topology and device \p dv_ind must match the local machine.
+ * I/O devices detection and the ROCm SMI component must be enabled in the
+ * topology. If not, the locality of the object may still be found using
+ * hwloc_rsmi_get_device_cpuset().
+ *
+ * \note The corresponding hwloc PCI device may be found by looking
+ * at the result parent pointer (unless PCI devices are filtered out).
+ */
+static __hwloc_inline hwloc_obj_t
+hwloc_rsmi_get_device_osdev(hwloc_topology_t topology, uint32_t dv_ind)
+{
+  hwloc_obj_t osdev;
+  rsmi_status_t ret;
+  uint64_t bdfid = 0;
+  unsigned domain, device, bus, func;
+  uint64_t id;
+  char uuid[64];
+
+  if (!hwloc_topology_is_thissystem(topology)) {
+    errno = EINVAL;
+    return NULL;
+  }
+
+  ret = rsmi_dev_pci_id_get(dv_ind, &bdfid);
+  if (RSMI_STATUS_SUCCESS != ret) {
+    errno = EINVAL;
+    return NULL;
+  }
+  domain = (bdfid>>32) & 0xffffffff;
+  bus = ((bdfid & 0xffff)>>8) & 0xff;
+  device = ((bdfid & 0xff)>>3) & 0x1f;
+  func = bdfid & 0x7;
+
+  ret = rsmi_dev_unique_id_get(dv_ind, &id);
+  if (RSMI_STATUS_SUCCESS != ret)
+    uuid[0] = '\0';
+  else
+    sprintf(uuid, "%lx", id);
+
+  osdev = NULL;
+  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
+    hwloc_obj_t pcidev = osdev->parent;
+    const char *info;
+
+    if (strncmp(osdev->name, "rsmi", 4))
+      continue;
+
+    if (pcidev
+      && pcidev->type == HWLOC_OBJ_PCI_DEVICE
+      && pcidev->attr->pcidev.domain == domain
+      && pcidev->attr->pcidev.bus == bus
+      && pcidev->attr->pcidev.dev == device
+      && pcidev->attr->pcidev.func == func)
+      return osdev;
+
+    info = hwloc_obj_get_info_by_name(osdev, "AMDUUID");
+    if (info && !strcmp(info, uuid))
+      return osdev;
+  }
+
+  return NULL;
+}
+
+/** @} */
+
+
+#ifdef __cplusplus
+} /* extern "C" */
+#endif
+
+
+#endif /* HWLOC_RSMI_H */

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

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2013-2018 Inria.  All rights reserved.
+ * Copyright © 2013-2023 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -48,6 +48,8 @@ extern "C" {
  * This length (in bytes) must be used in hwloc_shmem_topology_write()
  * and hwloc_shmem_topology_adopt() later.
  *
+ * \return the length, or -1 on error, for instance if flags are invalid.
+ *
  * \note Flags \p flags are currently unused, must be 0.
  */
 HWLOC_DECLSPEC int hwloc_shmem_topology_get_length(hwloc_topology_t topology,
@@ -74,9 +76,10 @@ HWLOC_DECLSPEC int hwloc_shmem_topology_get_length(hwloc_topology_t topology,
  * is not. However the caller may also allocate it manually in shared memory
  * to share it as well.
  *
- * \return -1 with errno set to EBUSY if the virtual memory mapping defined
+ * \return 0 on success.
+ * \return -1 with errno set to \c EBUSY if the virtual memory mapping defined
  * by \p mmap_address and \p length isn't available in the process.
- * \return -1 with errno set to EINVAL if \p fileoffset, \p mmap_address
+ * \return -1 with errno set to \c EINVAL if \p fileoffset, \p mmap_address
  * or \p length aren't page-aligned.
  */
 HWLOC_DECLSPEC int hwloc_shmem_topology_write(hwloc_topology_t topology,
@@ -112,14 +115,16 @@ HWLOC_DECLSPEC int hwloc_shmem_topology_write(hwloc_topology_t topology,
  *
  * \note This function takes care of calling hwloc_topology_abi_check().
  *
- * \return -1 with errno set to EBUSY if the virtual memory mapping defined
+ * \return 0 on success.
+ *
+ * \return -1 with errno set to \c EBUSY if the virtual memory mapping defined
  * by \p mmap_address and \p length isn't available in the process.
  *
- * \return -1 with errno set to EINVAL if \p fileoffset, \p mmap_address
+ * \return -1 with errno set to \c EINVAL if \p fileoffset, \p mmap_address
  * or \p length aren't page-aligned, or do not match what was given to
  * hwloc_shmem_topology_write() earlier.
  *
- * \return -1 with errno set to EINVAL if the layout of the topology structure
+ * \return -1 with errno set to \c EINVAL if the layout of the topology structure
  * is different between the writer process and the adopter process.
  */
 HWLOC_DECLSPEC int hwloc_shmem_topology_adopt(hwloc_topology_t *topologyp,

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

@@ -0,0 +1,76 @@
+/*
+ * Copyright © 2021 Inria.  All rights reserved.
+ * See COPYING in top-level directory.
+ */
+
+/** \file
+ * \brief Macros to help interaction between hwloc and Windows.
+ *
+ * Applications that use hwloc on Windows may want to include this file
+ * for Windows specific hwloc features.
+ */
+
+#ifndef HWLOC_WINDOWS_H
+#define HWLOC_WINDOWS_H
+
+#include "hwloc.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+/** \defgroup hwlocality_windows Windows-specific helpers
+ *
+ * These functions query Windows processor groups.
+ * These groups partition the operating system into virtual sets
+ * of up to 64 neighbor PUs.
+ * Threads and processes may only be bound inside a single group.
+ * Although Windows processor groups may be exposed in the hwloc
+ * hierarchy as hwloc Groups, they are also often merged into
+ * existing hwloc objects such as NUMA nodes or Packages.
+ * This API provides explicit information about Windows processor
+ * groups so that applications know whether binding to a large
+ * set of PUs may fail because it spans over multiple Windows
+ * processor groups.
+ *
+ * @{
+ */
+
+
+/** \brief Get the number of Windows processor groups
+ *
+ * \p flags must be 0 for now.
+ *
+ * \return at least \c 1 on success.
+ * \return -1 on error, for instance if the topology does not match
+ * the current system (e.g. loaded from another machine through XML).
+ */
+HWLOC_DECLSPEC int hwloc_windows_get_nr_processor_groups(hwloc_topology_t topology, unsigned long flags);
+
+/** \brief Get the CPU-set of a Windows processor group.
+ *
+ * Get the set of PU included in the processor group specified
+ * by \p pg_index.
+ * \p pg_index must be between \c 0 and the value returned
+ * by hwloc_windows_get_nr_processor_groups() minus 1.
+ *
+ * \p flags must be 0 for now.
+ *
+ * \return \c 0 on success.
+ * \return \c -1 on error, for instance if \p pg_index is invalid,
+ * or if the topology does not match the current system (e.g. loaded
+ * from another machine through XML).
+ */
+HWLOC_DECLSPEC int hwloc_windows_get_processor_group_cpuset(hwloc_topology_t topology, unsigned pg_index, hwloc_cpuset_t cpuset, unsigned long flags);
+
+/** @} */
+
+
+#ifdef __cplusplus
+} /* extern "C" */
+#endif
+
+
+#endif /* HWLOC_WINDOWS_H */

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

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

src/3rdparty/hwloc/include/private/cpuid-x86.h

@@ -11,6 +11,22 @@
 #ifndef HWLOC_PRIVATE_CPUID_X86_H
 #define HWLOC_PRIVATE_CPUID_X86_H
 
+/* A macro for annotating memory as uninitialized when building with MSAN
+ * (and otherwise having no effect). See below for why this is used with
+ * our custom assembly.
+ */
+#ifdef __has_feature
+#define HWLOC_HAS_FEATURE(name) __has_feature(name)
+#else
+#define HWLOC_HAS_FEATURE(name) 0
+#endif
+#if HWLOC_HAS_FEATURE(memory_sanitizer) || defined(MEMORY_SANITIZER)
+#include <sanitizer/msan_interface.h>
+#define HWLOC_ANNOTATE_MEMORY_IS_INITIALIZED(ptr, len) __msan_unpoison(ptr, len)
+#else
+#define HWLOC_ANNOTATE_MEMORY_IS_INITIALIZED(ptr, len)
+#endif
+
 #if (defined HWLOC_X86_32_ARCH) && (!defined HWLOC_HAVE_MSVC_CPUIDEX)
 static __hwloc_inline int hwloc_have_x86_cpuid(void)
 {
@@ -71,12 +87,18 @@ static __hwloc_inline void hwloc_x86_cpuid(unsigned *eax, unsigned *ebx, unsigne
   "movl %k2,%1\n\t"
   : "+a" (*eax), "=m" (*ebx), "=&r"(sav_rbx),
     "+c" (*ecx), "=&d" (*edx));
+  /* MSAN does not recognize the effect of the above assembly on the memory operand
+   * (`"=m"(*ebx)`). This may get improved in MSAN at some point in the future, e.g.
+   * see https://github.com/llvm/llvm-project/pull/77393. */
+  HWLOC_ANNOTATE_MEMORY_IS_INITIALIZED(ebx, sizeof *ebx);
 #elif defined(HWLOC_X86_32_ARCH)
   __asm__(
   "mov %%ebx,%1\n\t"
   "cpuid\n\t"
   "xchg %%ebx,%1\n\t"
   : "+a" (*eax), "=&SD" (*ebx), "+c" (*ecx), "=&d" (*edx));
+  /* See above. */
+  HWLOC_ANNOTATE_MEMORY_IS_INITIALIZED(ebx, sizeof *ebx);
 #else
 #error unknown architecture
 #endif

src/3rdparty/hwloc/include/private/internal-components.h

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2018-2019 Inria.  All rights reserved.
+ * Copyright © 2018-2020 Inria.  All rights reserved.
  *
  * See COPYING in top-level directory.
  */
@@ -31,6 +31,7 @@ HWLOC_DECLSPEC extern const struct hwloc_component hwloc_cuda_component;
 HWLOC_DECLSPEC extern const struct hwloc_component hwloc_gl_component;
 HWLOC_DECLSPEC extern const struct hwloc_component hwloc_nvml_component;
 HWLOC_DECLSPEC extern const struct hwloc_component hwloc_rsmi_component;
+HWLOC_DECLSPEC extern const struct hwloc_component hwloc_levelzero_component;
 HWLOC_DECLSPEC extern const struct hwloc_component hwloc_opencl_component;
 HWLOC_DECLSPEC extern const struct hwloc_component hwloc_pci_component;
 

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2019 Inria.  All rights reserved.
+ * Copyright © 2009-2024 Inria.  All rights reserved.
  * Copyright © 2009-2012 Université Bordeaux
  * Copyright © 2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -186,7 +186,7 @@ hwloc_ffsl_from_ffs32(unsigned long x)
 /**
  * flsl helpers.
  */
-#ifdef __GNUC_____
+#ifdef __GNUC__
 
 #  if (__GNUC__ >= 4) || ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))
 #    define hwloc_flsl(x) ((x) ? (8*sizeof(long) - __builtin_clzl(x)) : 0)
@@ -504,7 +504,7 @@ hwloc__obj_type_is_icache(hwloc_obj_type_t type)
   }                                    \
 } while(0)
 #else /* HAVE_USELOCALE */
-#if __HWLOC_HAVE_ATTRIBUTE_UNUSED
+#if HWLOC_HAVE_ATTRIBUTE_UNUSED
 #define hwloc_localeswitch_declare int __dummy_nolocale __hwloc_attribute_unused
 #define hwloc_localeswitch_init()
 #else
@@ -573,4 +573,35 @@ typedef SSIZE_T ssize_t;
 #  endif
 #endif
 
+static __inline float
+hwloc__pci_link_speed(unsigned generation, unsigned lanes)
+{
+  float lanespeed;
+  /*
+   * These are single-direction bandwidths only.
+   *
+   * Gen1 used NRZ with 8/10 encoding.
+   * PCIe Gen1 = 2.5GT/s signal-rate per lane x 8/10    =  0.25GB/s data-rate per lane
+   * PCIe Gen2 = 5  GT/s signal-rate per lane x 8/10    =  0.5 GB/s data-rate per lane
+   * Gen3 switched to NRZ with 128/130 encoding.
+   * PCIe Gen3 = 8  GT/s signal-rate per lane x 128/130 =  1   GB/s data-rate per lane
+   * PCIe Gen4 = 16 GT/s signal-rate per lane x 128/130 =  2   GB/s data-rate per lane
+   * PCIe Gen5 = 32 GT/s signal-rate per lane x 128/130 =  4   GB/s data-rate per lane
+   * Gen6 switched to PAM with with 242/256 FLIT (242B payload protected by 8B CRC + 6B FEC).
+   * PCIe Gen6 = 64 GT/s signal-rate per lane x 242/256 =  8   GB/s data-rate per lane
+   * PCIe Gen7 = 128GT/s signal-rate per lane x 242/256 = 16   GB/s data-rate per lane
+   */
+
+  /* lanespeed in Gbit/s */
+  if (generation <= 2)
+    lanespeed = 2.5f * generation * 0.8f;
+  else if (generation <= 5)
+    lanespeed = 8.0f * (1<<(generation-3)) * 128/130;
+  else
+    lanespeed = 8.0f * (1<<(generation-3)) * 242/256; /* assume Gen8 will be 256 GT/s and so on */
+
+  /* linkspeed in GB/s */
+  return lanespeed * lanes / 8;
+}
+
 #endif /* HWLOC_PRIVATE_MISC_H */

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

@@ -1,578 +0,0 @@
-/*
- * Copyright © 2014 Cisco Systems, Inc.  All rights reserved.
- * Copyright © 2013-2014 University of Wisconsin-La Crosse.
- *                         All rights reserved.
- * Copyright © 2015-2017 Inria.  All rights reserved.
- *
- * $COPYRIGHT$
- *
- * Additional copyrights may follow
- * See COPYING in top-level directory.
- *
- * $HEADER$
- */
-
-#ifndef _NETLOC_PRIVATE_H_
-#define _NETLOC_PRIVATE_H_
-
-#include <hwloc.h>
-#include <netloc.h>
-#include <netloc/uthash.h>
-#include <netloc/utarray.h>
-#include <private/autogen/config.h>
-
-#define NETLOCFILE_VERSION 1
-
-#ifdef NETLOC_SCOTCH
-#include <stdint.h>
-#include <scotch.h>
-#define NETLOC_int SCOTCH_Num
-#else
-#define NETLOC_int int
-#endif
-
-/*
- * "Import" a few things from hwloc
- */
-#define __netloc_attribute_unused __hwloc_attribute_unused
-#define __netloc_attribute_malloc __hwloc_attribute_malloc
-#define __netloc_attribute_const __hwloc_attribute_const
-#define __netloc_attribute_pure __hwloc_attribute_pure
-#define __netloc_attribute_deprecated __hwloc_attribute_deprecated
-#define __netloc_attribute_may_alias __hwloc_attribute_may_alias
-#define NETLOC_DECLSPEC HWLOC_DECLSPEC
-
-
-/**********************************************************************
- * Types
- **********************************************************************/
-
-/**
- * Definitions for Comparators
- * \sa These are the return values from the following functions:
- *     netloc_network_compare, netloc_dt_edge_t_compare, netloc_dt_node_t_compare
- */
-typedef enum {
-    NETLOC_CMP_SAME    =  0,  /**< Compared as the Same */
-    NETLOC_CMP_SIMILAR = -1,  /**< Compared as Similar, but not the Same */
-    NETLOC_CMP_DIFF    = -2   /**< Compared as Different */
-} netloc_compare_type_t;
-
-/**
- * Enumerated type for the various types of supported networks
- */
-typedef enum {
-    NETLOC_NETWORK_TYPE_ETHERNET    = 1, /**< Ethernet network */
-    NETLOC_NETWORK_TYPE_INFINIBAND  = 2, /**< InfiniBand network */
-    NETLOC_NETWORK_TYPE_INVALID     = 3  /**< Invalid network */
-} netloc_network_type_t;
-
-/**
- * Enumerated type for the various types of supported topologies
- */
-typedef enum {
-    NETLOC_TOPOLOGY_TYPE_INVALID = -1, /**< Invalid */
-    NETLOC_TOPOLOGY_TYPE_TREE    = 1,  /**< Tree */
-} netloc_topology_type_t;
-
-/**
- * Enumerated type for the various types of nodes
- */
-typedef enum {
-    NETLOC_NODE_TYPE_HOST    = 0, /**< Host (a.k.a., network addressable endpoint - e.g., MAC Address) node */
-    NETLOC_NODE_TYPE_SWITCH  = 1, /**< Switch node */
-    NETLOC_NODE_TYPE_INVALID = 2  /**< Invalid node */
-} netloc_node_type_t;
-
-typedef enum {
-    NETLOC_ARCH_TREE    =  0,  /* Fat tree */
-} netloc_arch_type_t;
-
-
-/* Pre declarations to avoid inter dependency problems */
-/** \cond IGNORE */
-struct netloc_topology_t;
-typedef struct netloc_topology_t netloc_topology_t;
-struct netloc_node_t;
-typedef struct netloc_node_t netloc_node_t;
-struct netloc_edge_t;
-typedef struct netloc_edge_t netloc_edge_t;
-struct netloc_physical_link_t;
-typedef struct netloc_physical_link_t netloc_physical_link_t;
-struct netloc_path_t;
-typedef struct netloc_path_t netloc_path_t;
-
-struct netloc_arch_tree_t;
-typedef struct netloc_arch_tree_t netloc_arch_tree_t;
-struct netloc_arch_node_t;
-typedef struct netloc_arch_node_t netloc_arch_node_t;
-struct netloc_arch_node_slot_t;
-typedef struct netloc_arch_node_slot_t netloc_arch_node_slot_t;
-struct netloc_arch_t;
-typedef struct netloc_arch_t netloc_arch_t;
-/** \endcond */
-
-/**
- * \struct netloc_topology_t
- * \brief Netloc Topology Context
- *
- * An opaque data structure used to reference a network topology.
- *
- * \note Must be initialized with \ref netloc_topology_construct()
- */
-struct netloc_topology_t {
-    /** Topology path */
-    char *topopath;
-    /** Subnet ID */
-    char *subnet_id;
-
-    /** Node List */
-    netloc_node_t *nodes; /* Hash table of nodes by physical_id */
-    netloc_node_t *nodesByHostname; /* Hash table of nodes by hostname */
-
-    netloc_physical_link_t *physical_links; /* Hash table with physcial links */
-
-    /** Partition List */
-    UT_array *partitions;
-
-    /** Hwloc topology List */
-    char *hwlocpath;
-    UT_array *topos;
-    hwloc_topology_t *hwloc_topos;
-
-    /** Type of the graph */
-    netloc_topology_type_t type;
-};
-
-/**
- * \brief Netloc Node Type
- *
- * Represents the concept of a node (a.k.a., vertex, endpoint) within a network
- * graph. This could be a server or a network switch. The \ref node_type parameter
- * will distinguish the exact type of node this represents in the graph.
- */
-struct netloc_node_t {
-    UT_hash_handle hh;       /* makes this structure hashable with physical_id */
-    UT_hash_handle hh2;      /* makes this structure hashable with hostname */
-
-    /** Physical ID of the node */
-    char physical_id[20];
-
-    /** Logical ID of the node (if any) */
-    int logical_id;
-
-    /** Type of the node */
-    netloc_node_type_t type;
-
-    /* Pointer to physical_links */
-    UT_array *physical_links;
-
-    /** Description information from discovery (if any) */
-    char *description;
-
-    /**
-     * Application-given private data pointer.
-     * Initialized to NULL, and not used by the netloc library.
-     */
-    void * userdata;
-
-    /** Outgoing edges from this node */
-    netloc_edge_t *edges;
-
-    UT_array *subnodes; /* the group of nodes for the virtual nodes */
-
-    netloc_path_t *paths;
-
-    char *hostname;
-
-    UT_array *partitions; /* index in the list from the topology */
-
-    hwloc_topology_t hwlocTopo;
-    int hwlocTopoIdx;
-};
-
-/**
- * \brief Netloc Edge Type
- *
- * Represents the concept of a directed edge within a network graph.
- *
- * \note We do not point to the netloc_node_t structure directly to
- * simplify the representation, and allow the information to more easily
- * be entered into the data store without circular references.
- * \todo JJH Is the note above still true?
- */
-struct netloc_edge_t {
-    UT_hash_handle hh;       /* makes this structure hashable */
-
-    netloc_node_t *dest;
-
-    int id;
-
-    /** Pointers to the parent node */
-    netloc_node_t *node;
-
-    /* Pointer to physical_links */
-    UT_array *physical_links;
-
-    /** total gbits of the links */
-    float total_gbits;
-
-    UT_array *partitions; /* index in the list from the topology */
-
-    UT_array *subnode_edges; /* for edges going to virtual nodes */
-
-    struct netloc_edge_t *other_way;
-
-    /**
-     * Application-given private data pointer.
-     * Initialized to NULL, and not used by the netloc library.
-     */
-    void * userdata;
-};
-
-
-struct netloc_physical_link_t {
-    UT_hash_handle hh;       /* makes this structure hashable */
-
-    int id; // TODO long long
-    netloc_node_t *src;
-    netloc_node_t *dest;
-    int ports[2];
-    char *width;
-    char *speed;
-
-    netloc_edge_t *edge;
-
-    int other_way_id;
-    struct netloc_physical_link_t *other_way;
-
-    UT_array *partitions; /* index in the list from the topology */
-
-    /** gbits of the link from speed and width */
-    float gbits;
-
-    /** Description information from discovery (if any) */
-    char *description;
-};
-
-struct netloc_path_t {
-    UT_hash_handle hh;       /* makes this structure hashable */
-    char dest_id[20];
-    UT_array *links;
-};
-
-
-/**********************************************************************
- *        Architecture structures
- **********************************************************************/
-struct netloc_arch_tree_t {
-    NETLOC_int num_levels;
-    NETLOC_int *degrees;
-    NETLOC_int *cost;
-};
-
-struct netloc_arch_node_t {
-    UT_hash_handle hh;       /* makes this structure hashable */
-    char *name; /* Hash key */
-    netloc_node_t *node; /* Corresponding node */
-    int idx_in_topo; /* idx with ghost hosts to have complete topo */
-    int num_slots; /* it is not the real number of slots but the maximum slot idx */
-    int *slot_idx; /* corresponding idx in slot_tree */
-    int *slot_os_idx; /* corresponding os index for each leaf in tree */
-    netloc_arch_tree_t *slot_tree; /* Tree built from hwloc */
-    int num_current_slots; /* Number of PUs */
-    NETLOC_int *current_slots; /* indices in the complete tree */
-    int *slot_ranks; /* corresponding MPI rank for each leaf in tree */
-};
-
-struct netloc_arch_node_slot_t {
-    netloc_arch_node_t *node;
-    int slot;
-};
-
-struct netloc_arch_t {
-    netloc_topology_t *topology;
-    int has_slots; /* if slots are included in the architecture */
-    netloc_arch_type_t type;
-    union {
-        netloc_arch_tree_t *node_tree;
-        netloc_arch_tree_t *global_tree;
-    } arch;
-    netloc_arch_node_t *nodes_by_name;
-    netloc_arch_node_slot_t *node_slot_by_idx; /* node_slot by index in complete topo */
-    NETLOC_int num_current_hosts; /* if has_slots, host is a slot, else host is a node */
-    NETLOC_int *current_hosts; /* indices in the complete topology */
-};
-
-/**********************************************************************
- * Topology Functions
- **********************************************************************/
-/**
- * Allocate a topology handle.
- *
- * User is responsible for calling \ref netloc_detach on the topology handle.
- * The network parameter information is deep copied into the topology handle, so the
- * user may destruct the network handle after calling this function and/or reuse
- * the network handle.
- *
- * \returns NETLOC_SUCCESS on success
- * \returns NETLOC_ERROR upon an error.
- */
-netloc_topology_t *netloc_topology_construct(char *path);
-
-/**
- * Destruct a topology handle
- *
- * \param topology A valid pointer to a \ref netloc_topology_t handle created
- * from a prior call to \ref netloc_topology_construct.
- *
- * \returns NETLOC_SUCCESS on success
- * \returns NETLOC_ERROR upon an error.
- */
-int netloc_topology_destruct(netloc_topology_t *topology);
-
-int netloc_topology_find_partition_idx(netloc_topology_t *topology, char *partition_name);
-
-int netloc_topology_read_hwloc(netloc_topology_t *topology, int num_nodes,
-        netloc_node_t **node_list);
-
-#define netloc_topology_iter_partitions(topology,partition) \
-    for ((partition) = (char **)utarray_front(topology->partitions); \
-            (partition) != NULL; \
-            (partition) = (char **)utarray_next(topology->partitions, partition))
-
-#define netloc_topology_iter_hwloctopos(topology,hwloctopo) \
-    for ((hwloctopo) = (char **)utarray_front(topology->topos); \
-            (hwloctopo) != NULL; \
-            (hwloctopo) = (char **)utarray_next(topology->topos, hwloctopo))
-
-#define netloc_topology_find_node(topology,node_id,node) \
-    HASH_FIND_STR(topology->nodes, node_id, node)
-
-#define netloc_topology_iter_nodes(topology,node,_tmp) \
-    HASH_ITER(hh, topology->nodes, node, _tmp)
-
-#define netloc_topology_num_nodes(topology) \
-    HASH_COUNT(topology->nodes)
-
-/*************************************************/
-
-
-/**
- * Constructor for netloc_node_t
- *
- * User is responsible for calling the destructor on the handle.
- *
- * Returns
- *   A newly allocated pointer to the network information.
- */
-netloc_node_t *netloc_node_construct(void);
-
-/**
- * Destructor for netloc_node_t
- *
- * \param node A valid node handle
- *
- * Returns
- *   NETLOC_SUCCESS on success
- *   NETLOC_ERROR on error
- */
-int netloc_node_destruct(netloc_node_t *node);
-
-char *netloc_node_pretty_print(netloc_node_t* node);
-
-#define netloc_node_get_num_subnodes(node) \
-    utarray_len((node)->subnodes)
-
-#define netloc_node_get_subnode(node,i) \
-    (*(netloc_node_t **)utarray_eltptr((node)->subnodes, (i)))
-
-#define netloc_node_get_num_edges(node) \
-    utarray_len((node)->edges)
-
-#define netloc_node_get_edge(node,i) \
-    (*(netloc_edge_t **)utarray_eltptr((node)->edges, (i)))
-
-#define netloc_node_iter_edges(node,edge,_tmp) \
-    HASH_ITER(hh, node->edges, edge, _tmp)
-
-#define netloc_node_iter_paths(node,path,_tmp) \
-    HASH_ITER(hh, node->paths, path, _tmp)
-
-#define netloc_node_is_host(node) \
-    (node->type == NETLOC_NODE_TYPE_HOST)
-
-#define netloc_node_is_switch(node) \
-    (node->type == NETLOC_NODE_TYPE_SWITCH)
-
-#define netloc_node_iter_paths(node, path,_tmp) \
-    HASH_ITER(hh, node->paths, path, _tmp)
-
-int netloc_node_is_in_partition(netloc_node_t *node, int partition);
-
-/*************************************************/
-
-
-/**
- * Constructor for netloc_edge_t
- *
- * User is responsible for calling the destructor on the handle.
- *
- * Returns
- *   A newly allocated pointer to the edge information.
- */
-netloc_edge_t *netloc_edge_construct(void);
-
-/**
- * Destructor for netloc_edge_t
- *
- * \param edge A valid edge handle
- *
- * Returns
- *   NETLOC_SUCCESS on success
- *   NETLOC_ERROR on error
- */
-int netloc_edge_destruct(netloc_edge_t *edge);
-
-char * netloc_edge_pretty_print(netloc_edge_t* edge);
-
-void netloc_edge_reset_uid(void);
-
-int netloc_edge_is_in_partition(netloc_edge_t *edge, int partition);
-
-#define netloc_edge_get_num_links(edge) \
-    utarray_len((edge)->physical_links)
-
-#define netloc_edge_get_link(edge,i) \
-    (*(netloc_physical_link_t **)utarray_eltptr((edge)->physical_links, (i)))
-
-#define netloc_edge_get_num_subedges(edge) \
-    utarray_len((edge)->subnode_edges)
-
-#define netloc_edge_get_subedge(edge,i) \
-    (*(netloc_edge_t **)utarray_eltptr((edge)->subnode_edges, (i)))
-
-/*************************************************/
-
-
-/**
- * Constructor for netloc_physical_link_t
- *
- * User is responsible for calling the destructor on the handle.
- *
- * Returns
- *   A newly allocated pointer to the physical link information.
- */
-netloc_physical_link_t * netloc_physical_link_construct(void);
-
-/**
- * Destructor for netloc_physical_link_t
- *
- * Returns
- *   NETLOC_SUCCESS on success
- *   NETLOC_ERROR on error
- */
-int netloc_physical_link_destruct(netloc_physical_link_t *link);
-
-char * netloc_link_pretty_print(netloc_physical_link_t* link);
-
-/*************************************************/
-
-
-netloc_path_t *netloc_path_construct(void);
-int netloc_path_destruct(netloc_path_t *path);
-
-
-/**********************************************************************
- *        Architecture functions
- **********************************************************************/
-
-netloc_arch_t * netloc_arch_construct(void);
-
-int netloc_arch_destruct(netloc_arch_t *arch);
-
-int netloc_arch_build(netloc_arch_t *arch, int add_slots);
-
-int netloc_arch_set_current_resources(netloc_arch_t *arch);
-
-int netloc_arch_set_global_resources(netloc_arch_t *arch);
-
-int netloc_arch_node_get_hwloc_info(netloc_arch_node_t *arch);
-
-void netloc_arch_tree_complete(netloc_arch_tree_t *tree, UT_array **down_degrees_by_level,
-        int num_hosts, int **parch_idx);
-
-NETLOC_int netloc_arch_tree_num_leaves(netloc_arch_tree_t *tree);
-
-
-/**********************************************************************
- *        Access functions of various elements of the topology
- **********************************************************************/
-
-#define netloc_get_num_partitions(object) \
-    utarray_len((object)->partitions)
-
-#define netloc_get_partition(object,i) \
-    (*(int *)utarray_eltptr((object)->partitions, (i)))
-
-
-#define netloc_path_iter_links(path,link) \
-    for ((link) = (netloc_physical_link_t **)utarray_front(path->links); \
-            (link) != NULL; \
-            (link) = (netloc_physical_link_t **)utarray_next(path->links, link))
-
-/**********************************************************************
- *        Misc functions
- **********************************************************************/
-
-/**
- * Decode the network type
- *
- * \param net_type A valid member of the \ref netloc_network_type_t type
- *
- * \returns NULL if the type is invalid
- * \returns A string for that \ref netloc_network_type_t type
- */
-static inline const char * netloc_network_type_decode(netloc_network_type_t net_type) {
-    if( NETLOC_NETWORK_TYPE_ETHERNET == net_type ) {
-        return "ETH";
-    }
-    else if( NETLOC_NETWORK_TYPE_INFINIBAND == net_type ) {
-        return "IB";
-    }
-    else {
-        return NULL;
-    }
-}
-
-/**
- * Decode the node type
- *
- * \param node_type A valid member of the \ref netloc_node_type_t type
- *
- * \returns NULL if the type is invalid
- * \returns A string for that \ref netloc_node_type_t type
- */
-static inline const char * netloc_node_type_decode(netloc_node_type_t node_type) {
-    if( NETLOC_NODE_TYPE_SWITCH == node_type ) {
-        return "SW";
-    }
-    else if( NETLOC_NODE_TYPE_HOST == node_type ) {
-        return "CA";
-    }
-    else {
-        return NULL;
-    }
-}
-
-ssize_t netloc_line_get(char **lineptr, size_t *n, FILE *stream);
-
-char *netloc_line_get_next_token(char **string, char c);
-
-int netloc_build_comm_mat(char *filename, int *pn, double ***pmat);
-
-#define STRDUP_IF_NOT_NULL(str) (NULL == str ? NULL : strdup(str))
-#define STR_EMPTY_IF_NULL(str) (NULL == str ? "" : str)
-
-
-#endif // _NETLOC_PRIVATE_H_

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009      CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2025 Inria.  All rights reserved.
  * Copyright © 2009-2012, 2020 Université Bordeaux
  * Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
  *
@@ -166,6 +166,7 @@ struct hwloc_topology {
     unsigned long kind;
 
 #define HWLOC_INTERNAL_DIST_FLAG_OBJS_VALID (1U<<0) /* if the objs array is valid below */
+#define HWLOC_INTERNAL_DIST_FLAG_NOT_COMMITTED (1U<<1) /* if the distances isn't in the list yet */
     unsigned iflags;
 
     /* objects are currently stored in physical_index order */
@@ -244,6 +245,12 @@ struct hwloc_topology {
    * temporary variables during discovery
    */
 
+  /* set to 1 at the beginning of load() if the filter of any cpu cache type (L1 to L3i) is not NONE,
+   * may be checked by backends before querying caches
+   * (when they don't know the level of caches they are querying).
+   */
+  int want_some_cpu_caches;
+
   /* machine-wide memory.
    * temporarily stored there by OSes that only provide this without NUMA information,
    * and actually used later by the core.
@@ -258,6 +265,7 @@ struct hwloc_topology {
     unsigned bus_first, bus_last;
     hwloc_bitmap_t cpuset;
   } * pci_forced_locality;
+  hwloc_uint64_t pci_locality_quirks;
 
   /* component blacklisting */
   unsigned nr_blacklisted_components;
@@ -294,6 +302,9 @@ extern void hwloc__reorder_children(hwloc_obj_t parent);
 extern void hwloc_topology_setup_defaults(struct hwloc_topology *topology);
 extern void hwloc_topology_clear(struct hwloc_topology *topology);
 
+#define _HWLOC_RECONNECT_FLAG_KEEPSTRUCTURE (1UL<<0)
+extern int hwloc__reconnect(struct hwloc_topology *topology, unsigned long flags);
+
 /* insert memory object as memory child of normal parent */
 extern struct hwloc_obj * hwloc__attach_memory_object(struct hwloc_topology *topology, hwloc_obj_t parent,
                                                       hwloc_obj_t obj, const char *reason);
@@ -304,11 +315,6 @@ extern void hwloc_pci_discovery_init(struct hwloc_topology *topology);
 extern void hwloc_pci_discovery_prepare(struct hwloc_topology *topology);
 extern void hwloc_pci_discovery_exit(struct hwloc_topology *topology);
 
-/* Look for an object matching the given domain/bus/func,
- * either exactly or return the smallest container bridge
- */
-extern struct hwloc_obj * hwloc_pci_find_by_busid(struct hwloc_topology *topology, unsigned domain, unsigned bus, unsigned dev, unsigned func);
-
 /* Look for an object matching complete cpuset exactly, or insert one.
  * Return NULL on failure.
  * Return a good fallback (object above) on failure to insert.
@@ -408,10 +414,14 @@ extern void hwloc_internal_distances_prepare(hwloc_topology_t topology);
 extern void hwloc_internal_distances_destroy(hwloc_topology_t topology);
 extern int hwloc_internal_distances_dup(hwloc_topology_t new, hwloc_topology_t old);
 extern void hwloc_internal_distances_refresh(hwloc_topology_t topology);
-extern int hwloc_internal_distances_add(hwloc_topology_t topology, const char *name, unsigned nbobjs, hwloc_obj_t *objs, uint64_t *values, unsigned long kind, unsigned long flags);
-extern int hwloc_internal_distances_add_by_index(hwloc_topology_t topology, const char *name, hwloc_obj_type_t unique_type, hwloc_obj_type_t *different_types, unsigned nbobjs, uint64_t *indexes, uint64_t *values, unsigned long kind, unsigned long flags);
 extern void hwloc_internal_distances_invalidate_cached_objs(hwloc_topology_t topology);
 
+/* these distances_add() functions are higher-level than those in hwloc/plugins.h
+ * but they may change in the future, hence they are not exported to plugins.
+ */
+extern int hwloc_internal_distances_add_by_index(hwloc_topology_t topology, const char *name, hwloc_obj_type_t unique_type, hwloc_obj_type_t *different_types, unsigned nbobjs, uint64_t *indexes, uint64_t *values, unsigned long kind, unsigned long flags);
+extern int hwloc_internal_distances_add(hwloc_topology_t topology, const char *name, unsigned nbobjs, hwloc_obj_t *objs, uint64_t *values, unsigned long kind, unsigned long flags);
+
 extern void hwloc_internal_memattrs_init(hwloc_topology_t topology);
 extern void hwloc_internal_memattrs_prepare(hwloc_topology_t topology);
 extern void hwloc_internal_memattrs_destroy(hwloc_topology_t topology);
@@ -419,6 +429,7 @@ extern void hwloc_internal_memattrs_need_refresh(hwloc_topology_t topology);
 extern void hwloc_internal_memattrs_refresh(hwloc_topology_t topology);
 extern int hwloc_internal_memattrs_dup(hwloc_topology_t new, hwloc_topology_t old);
 extern int hwloc_internal_memattr_set_value(hwloc_topology_t topology, hwloc_memattr_id_t id, hwloc_obj_type_t target_type, hwloc_uint64_t target_gp_index, unsigned target_os_index, struct hwloc_internal_location_s *initiator, hwloc_uint64_t value);
+extern int hwloc_internal_memattrs_guess_memory_tiers(hwloc_topology_t topology, int force_subtype);
 
 extern void hwloc_internal_cpukinds_init(hwloc_topology_t topology);
 extern int hwloc_internal_cpukinds_rank(hwloc_topology_t topology);
@@ -475,11 +486,13 @@ extern char * hwloc_progname(struct hwloc_topology *topology);
 #define HWLOC_GROUP_KIND_INTEL_DIE			104	/* no subkind */
 #define HWLOC_GROUP_KIND_S390_BOOK			110	/* subkind 0 is book, subkind 1 is drawer (group of books) */
 #define HWLOC_GROUP_KIND_AMD_COMPUTE_UNIT		120	/* no subkind */
+#define HWLOC_GROUP_KIND_AMD_COMPLEX                    121     /* no subkind */
 /* then, OS-specific groups */
 #define HWLOC_GROUP_KIND_SOLARIS_PG_HW_PERF		200	/* subkind is group width */
 #define HWLOC_GROUP_KIND_AIX_SDL_UNKNOWN		210	/* subkind is SDL level */
 #define HWLOC_GROUP_KIND_WINDOWS_PROCESSOR_GROUP	220	/* no subkind */
 #define HWLOC_GROUP_KIND_WINDOWS_RELATIONSHIP_UNKNOWN	221	/* no subkind */
+#define HWLOC_GROUP_KIND_LINUX_CLUSTER                  222     /* no subkind */
 /* distance groups */
 #define HWLOC_GROUP_KIND_DISTANCE			900	/* subkind is round of adding these groups during distance based grouping */
 /* finally, hwloc-specific groups required to insert something else, should disappear as soon as possible */

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

@@ -0,0 +1,30 @@
+/*
+ * Copyright © 2009 Université Bordeaux
+ * Copyright © 2020-2022 Inria.  All rights reserved.
+ *
+ * See COPYING in top-level directory.
+ */
+
+#ifndef HWLOC_PRIVATE_WINDOWS_H
+#define HWLOC_PRIVATE_WINDOWS_H
+
+#ifndef _ANONYMOUS_UNION
+#ifdef __GNUC__
+#define _ANONYMOUS_UNION __extension__
+#else
+#define _ANONYMOUS_UNION
+#endif /* __GNUC__ */
+#endif /* _ANONYMOUS_UNION */
+
+#ifndef _ANONYMOUS_STRUCT
+#ifdef __GNUC__
+#define _ANONYMOUS_STRUCT __extension__
+#else
+#define _ANONYMOUS_STRUCT
+#endif /* __GNUC__ */
+#endif /* _ANONYMOUS_STRUCT */
+
+#define DUMMYUNIONNAME
+#define DUMMYSTRUCTNAME
+
+#endif /* HWLOC_PRIVATE_WINDOWS_H */

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

@@ -19,13 +19,14 @@ HWLOC_DECLSPEC int hwloc__xml_verbose(void);
 typedef struct hwloc__xml_import_state_s {
   struct hwloc__xml_import_state_s *parent;
 
-  /* globals shared because the entire stack of states during import */
+  /* globals shared between the entire stack of states during import */
   struct hwloc_xml_backend_data_s *global;
 
   /* opaque data used to store backend-specific data.
    * statically allocated to allow stack-allocation by the common code without knowing actual backend needs.
+   * libxml is 3 ptrs. nolibxml is 3 ptr + one int.
    */
-  char data[32];
+  char data[4 * SIZEOF_VOID_P];
 } * hwloc__xml_import_state_t;
 
 struct hwloc__xml_imported_v1distances_s {
@@ -74,8 +75,9 @@ typedef struct hwloc__xml_export_state_s {
 
   /* opaque data used to store backend-specific data.
    * statically allocated to allow stack-allocation by the common code without knowing actual backend needs.
+   * libxml is 1 ptr. nolibxml is 1 ptr + 2 size_t + 3 ints.
    */
-  char data[40];
+  char data[6 * SIZEOF_VOID_P];
 } * hwloc__xml_export_state_t;
 
 HWLOC_DECLSPEC void hwloc__xml_export_topology(hwloc__xml_export_state_t parentstate, hwloc_topology_t topology, unsigned long flags);

src/3rdparty/hwloc/src/bind.c

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2024 Inria.  All rights reserved.
  * Copyright © 2009-2010, 2012 Université Bordeaux
  * Copyright © 2011-2015 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -287,6 +287,7 @@ static __hwloc_inline int hwloc__check_membind_policy(hwloc_membind_policy_t pol
       || policy == HWLOC_MEMBIND_FIRSTTOUCH
       || policy == HWLOC_MEMBIND_BIND
       || policy == HWLOC_MEMBIND_INTERLEAVE
+      || policy == HWLOC_MEMBIND_WEIGHTED_INTERLEAVE
       || policy == HWLOC_MEMBIND_NEXTTOUCH)
     return 0;
   return -1;

src/3rdparty/hwloc/src/bitmap.c

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2024 Inria.  All rights reserved.
  * Copyright © 2009-2011 Université Bordeaux
  * Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -245,6 +245,7 @@ int hwloc_bitmap_copy(struct hwloc_bitmap_s * dst, const struct hwloc_bitmap_s *
 /* Strings always use 32bit groups */
 #define HWLOC_PRIxSUBBITMAP		"%08lx"
 #define HWLOC_BITMAP_SUBSTRING_SIZE	32
+#define HWLOC_BITMAP_SUBSTRING_FULL_VALUE 0xFFFFFFFFUL
 #define HWLOC_BITMAP_SUBSTRING_LENGTH	(HWLOC_BITMAP_SUBSTRING_SIZE/4)
 #define HWLOC_BITMAP_STRING_PER_LONG	(HWLOC_BITS_PER_LONG/HWLOC_BITMAP_SUBSTRING_SIZE)
 
@@ -261,6 +262,7 @@ int hwloc_bitmap_snprintf(char * __hwloc_restrict buf, size_t buflen, const stru
   const unsigned long accum_mask = ~0UL;
 #else /* HWLOC_BITS_PER_LONG != HWLOC_BITMAP_SUBSTRING_SIZE */
   const unsigned long accum_mask = ((1UL << HWLOC_BITMAP_SUBSTRING_SIZE) - 1) << (HWLOC_BITS_PER_LONG - HWLOC_BITMAP_SUBSTRING_SIZE);
+  int merge_with_infinite_prefix = 0;
 #endif /* HWLOC_BITS_PER_LONG != HWLOC_BITMAP_SUBSTRING_SIZE */
 
   HWLOC__BITMAP_CHECK(set);
@@ -279,6 +281,9 @@ int hwloc_bitmap_snprintf(char * __hwloc_restrict buf, size_t buflen, const stru
       res = size>0 ? (int)size - 1 : 0;
     tmp += res;
     size -= res;
+#if HWLOC_BITS_PER_LONG > HWLOC_BITMAP_SUBSTRING_SIZE
+    merge_with_infinite_prefix = 1;
+#endif
   }
 
   i=(int) set->ulongs_count-1;
@@ -294,16 +299,24 @@ int hwloc_bitmap_snprintf(char * __hwloc_restrict buf, size_t buflen, const stru
   }
 
   while (i>=0 || accumed) {
+    unsigned long value;
+
     /* Refill accumulator */
     if (!accumed) {
       accum = set->ulongs[i--];
       accumed = HWLOC_BITS_PER_LONG;
     }
+    value = (accum & accum_mask) >> (HWLOC_BITS_PER_LONG - HWLOC_BITMAP_SUBSTRING_SIZE);
 
-    if (accum & accum_mask) {
+#if HWLOC_BITS_PER_LONG > HWLOC_BITMAP_SUBSTRING_SIZE
+    if (merge_with_infinite_prefix && value == HWLOC_BITMAP_SUBSTRING_FULL_VALUE) {
+      /* first full subbitmap merged with infinite prefix */
+      res = 0;
+    } else
+#endif
+    if (value) {
       /* print the whole subset if not empty */
-        res = hwloc_snprintf(tmp, size, needcomma ? ",0x" HWLOC_PRIxSUBBITMAP : "0x" HWLOC_PRIxSUBBITMAP,
+      res = hwloc_snprintf(tmp, size, needcomma ? ",0x" HWLOC_PRIxSUBBITMAP : "0x" HWLOC_PRIxSUBBITMAP, value);
-		     (accum & accum_mask) >> (HWLOC_BITS_PER_LONG - HWLOC_BITMAP_SUBSTRING_SIZE));
       needcomma = 1;
     } else if (i == -1 && accumed == HWLOC_BITMAP_SUBSTRING_SIZE) {
       /* print a single 0 to mark the last subset */
@@ -323,6 +336,7 @@ int hwloc_bitmap_snprintf(char * __hwloc_restrict buf, size_t buflen, const stru
 #else
     accum <<= HWLOC_BITMAP_SUBSTRING_SIZE;
     accumed -= HWLOC_BITMAP_SUBSTRING_SIZE;
+    merge_with_infinite_prefix = 0;
 #endif
 
     if (res >= size)
@@ -362,7 +376,8 @@ int hwloc_bitmap_sscanf(struct hwloc_bitmap_s *set, const char * __hwloc_restric
 {
   const char * current = string;
   unsigned long accum = 0;
-  int count=0;
+  int count = 0;
+  int ulongcount;
   int infinite = 0;
 
   /* count how many substrings there are */
@@ -383,9 +398,20 @@ int hwloc_bitmap_sscanf(struct hwloc_bitmap_s *set, const char * __hwloc_restric
     count--;
   }
 
-  if (hwloc_bitmap_reset_by_ulongs(set, (count + HWLOC_BITMAP_STRING_PER_LONG - 1) / HWLOC_BITMAP_STRING_PER_LONG) < 0)
+  ulongcount = (count + HWLOC_BITMAP_STRING_PER_LONG - 1) / HWLOC_BITMAP_STRING_PER_LONG;
+  if (hwloc_bitmap_reset_by_ulongs(set, ulongcount) < 0)
     return -1;
-  set->infinite = 0;
+
+  set->infinite = 0; /* will be updated later */
+
+#if HWLOC_BITS_PER_LONG != HWLOC_BITMAP_SUBSTRING_SIZE
+  if (infinite && (count % HWLOC_BITMAP_STRING_PER_LONG) != 0) {
+    /* accumulate substrings of the first ulong that are hidden in the infinite prefix */
+    int i;
+    for(i = (count % HWLOC_BITMAP_STRING_PER_LONG); i < HWLOC_BITMAP_STRING_PER_LONG; i++)
+      accum |= (HWLOC_BITMAP_SUBSTRING_FULL_VALUE << (i*HWLOC_BITMAP_SUBSTRING_SIZE));
+  }
+#endif
 
   while (*current != '\0') {
     unsigned long val;
@@ -544,6 +570,9 @@ int hwloc_bitmap_taskset_snprintf(char * __hwloc_restrict buf, size_t buflen, co
   ssize_t size = buflen;
   char *tmp = buf;
   int res, ret = 0;
+#if HWLOC_BITS_PER_LONG == 64
+  int merge_with_infinite_prefix = 0;
+#endif
   int started = 0;
   int i;
 
@@ -563,6 +592,9 @@ int hwloc_bitmap_taskset_snprintf(char * __hwloc_restrict buf, size_t buflen, co
       res = size>0 ? (int)size - 1 : 0;
     tmp += res;
     size -= res;
+#if HWLOC_BITS_PER_LONG == 64
+    merge_with_infinite_prefix = 1;
+#endif
   }
 
   i=set->ulongs_count-1;
@@ -582,7 +614,11 @@ int hwloc_bitmap_taskset_snprintf(char * __hwloc_restrict buf, size_t buflen, co
     if (started) {
       /* print the whole subset */
 #if HWLOC_BITS_PER_LONG == 64
-      res = hwloc_snprintf(tmp, size, "%016lx", val);
+      if (merge_with_infinite_prefix && (val & 0xffffffff00000000UL) == 0xffffffff00000000UL) {
+        res = hwloc_snprintf(tmp, size, "%08lx", val & 0xffffffffUL);
+      } else  {
+        res = hwloc_snprintf(tmp, size, "%016lx", val);
+      }
 #else
       res = hwloc_snprintf(tmp, size, "%08lx", val);
 #endif
@@ -599,6 +635,9 @@ int hwloc_bitmap_taskset_snprintf(char * __hwloc_restrict buf, size_t buflen, co
       res = size>0 ? (int)size - 1 : 0;
     tmp += res;
     size -= res;
+#if HWLOC_BITS_PER_LONG == 64
+    merge_with_infinite_prefix = 0;
+#endif
   }
 
   /* if didn't display anything, display 0x0 */
@@ -679,6 +718,10 @@ int hwloc_bitmap_taskset_sscanf(struct hwloc_bitmap_s *set, const char * __hwloc
       goto failed;
 
     set->ulongs[count-1] = val;
+    if (infinite && tmpchars != HWLOC_BITS_PER_LONG/4) {
+      /* infinite prefix with partial substring, fill remaining bits */
+      set->ulongs[count-1] |= (~0ULL)<<(4*tmpchars);
+    }
 
     current += tmpchars;
     chars -= tmpchars;

src/3rdparty/hwloc/src/components.c

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2022 Inria.  All rights reserved.
  * Copyright © 2012 Université Bordeaux
  * See COPYING in top-level directory.
  */
@@ -94,8 +94,7 @@ static hwloc_dlhandle hwloc_dlopenext(const char *_filename)
 {
   hwloc_dlhandle handle;
   char *filename = NULL;
-  (void) asprintf(&filename, "%s.so", _filename);
+  if (asprintf(&filename, "%s.so", _filename) < 0)
-  if (!filename)
     return NULL;
   handle = dlopen(filename, RTLD_NOW|RTLD_LOCAL);
   free(filename);
@@ -124,7 +123,7 @@ hwloc_dlforeachfile(const char *_paths,
       *colon = '\0';
 
     if (hwloc_plugins_verbose)
-      fprintf(stderr, " Looking under %s\n", path);
+      fprintf(stderr, "hwloc:  Looking under %s\n", path);
 
     dir = opendir(path);
     if (!dir)
@@ -198,7 +197,7 @@ hwloc__dlforeach_cb(const char *filename, void *_data __hwloc_attribute_unused)
   char *componentsymbolname;
 
   if (hwloc_plugins_verbose)
-    fprintf(stderr, "Plugin dlforeach found `%s'\n", filename);
+    fprintf(stderr, "hwloc: Plugin dlforeach found `%s'\n", filename);
 
   basename = strrchr(filename, '/');
   if (!basename)
@@ -208,7 +207,7 @@ hwloc__dlforeach_cb(const char *filename, void *_data __hwloc_attribute_unused)
 
   if (hwloc_plugins_blacklist && strstr(hwloc_plugins_blacklist, basename)) {
     if (hwloc_plugins_verbose)
-      fprintf(stderr, "Plugin `%s' is blacklisted in the environment\n", basename);
+      fprintf(stderr, "hwloc: Plugin `%s' is blacklisted in the environment\n", basename);
     goto out;
   }
 
@@ -216,14 +215,14 @@ hwloc__dlforeach_cb(const char *filename, void *_data __hwloc_attribute_unused)
   handle = hwloc_dlopenext(filename);
   if (!handle) {
     if (hwloc_plugins_verbose)
-      fprintf(stderr, "Failed to load plugin: %s\n", hwloc_dlerror());
+      fprintf(stderr, "hwloc: Failed to load plugin: %s\n", hwloc_dlerror());
     goto out;
   }
 
   componentsymbolname = malloc(strlen(basename)+10+1);
   if (!componentsymbolname) {
     if (hwloc_plugins_verbose)
-      fprintf(stderr, "Failed to allocation component `%s' symbol\n",
+      fprintf(stderr, "hwloc: Failed to allocation component `%s' symbol\n",
 	      basename);
     goto out_with_handle;
   }
@@ -231,38 +230,38 @@ hwloc__dlforeach_cb(const char *filename, void *_data __hwloc_attribute_unused)
   component = hwloc_dlsym(handle, componentsymbolname);
   if (!component) {
     if (hwloc_plugins_verbose)
-      fprintf(stderr, "Failed to find component symbol `%s'\n",
+      fprintf(stderr, "hwloc: Failed to find component symbol `%s'\n",
 	      componentsymbolname);
     free(componentsymbolname);
     goto out_with_handle;
   }
   if (component->abi != HWLOC_COMPONENT_ABI) {
     if (hwloc_plugins_verbose)
-      fprintf(stderr, "Plugin symbol ABI %u instead of %d\n",
+      fprintf(stderr, "hwloc: Plugin symbol ABI %u instead of %d\n",
 	      component->abi, HWLOC_COMPONENT_ABI);
     free(componentsymbolname);
     goto out_with_handle;
   }
   if (hwloc_plugins_verbose)
-    fprintf(stderr, "Plugin contains expected symbol `%s'\n",
+    fprintf(stderr, "hwloc: Plugin contains expected symbol `%s'\n",
 	    componentsymbolname);
   free(componentsymbolname);
 
   if (HWLOC_COMPONENT_TYPE_DISC == component->type) {
     if (strncmp(basename, "hwloc_", 6)) {
       if (hwloc_plugins_verbose)
-	fprintf(stderr, "Plugin name `%s' doesn't match its type DISCOVERY\n", basename);
+	fprintf(stderr, "hwloc: Plugin name `%s' doesn't match its type DISCOVERY\n", basename);
       goto out_with_handle;
     }
   } else if (HWLOC_COMPONENT_TYPE_XML == component->type) {
     if (strncmp(basename, "hwloc_xml_", 10)) {
       if (hwloc_plugins_verbose)
-	fprintf(stderr, "Plugin name `%s' doesn't match its type XML\n", basename);
+	fprintf(stderr, "hwloc: Plugin name `%s' doesn't match its type XML\n", basename);
       goto out_with_handle;
     }
   } else {
     if (hwloc_plugins_verbose)
-      fprintf(stderr, "Plugin name `%s' has invalid type %u\n",
+      fprintf(stderr, "hwloc: Plugin name `%s' has invalid type %u\n",
 	      basename, (unsigned) component->type);
     goto out_with_handle;
   }
@@ -277,7 +276,7 @@ hwloc__dlforeach_cb(const char *filename, void *_data __hwloc_attribute_unused)
   desc->handle = handle;
   desc->next = NULL;
   if (hwloc_plugins_verbose)
-    fprintf(stderr, "Plugin descriptor `%s' ready\n", basename);
+    fprintf(stderr, "hwloc: Plugin descriptor `%s' ready\n", basename);
 
   /* append to the list */
   prevdesc = &hwloc_plugins;
@@ -285,7 +284,7 @@ hwloc__dlforeach_cb(const char *filename, void *_data __hwloc_attribute_unused)
     prevdesc = &((*prevdesc)->next);
   *prevdesc = desc;
   if (hwloc_plugins_verbose)
-    fprintf(stderr, "Plugin descriptor `%s' queued\n", basename);
+    fprintf(stderr, "hwloc: Plugin descriptor `%s' queued\n", basename);
   return 0;
 
  out_with_handle:
@@ -300,7 +299,7 @@ hwloc_plugins_exit(void)
   struct hwloc__plugin_desc *desc, *next;
 
   if (hwloc_plugins_verbose)
-    fprintf(stderr, "Closing all plugins\n");
+    fprintf(stderr, "hwloc: Closing all plugins\n");
 
   desc = hwloc_plugins;
   while (desc) {
@@ -340,7 +339,7 @@ hwloc_plugins_init(void)
   hwloc_plugins = NULL;
 
   if (hwloc_plugins_verbose)
-    fprintf(stderr, "Starting plugin dlforeach in %s\n", path);
+    fprintf(stderr, "hwloc: Starting plugin dlforeach in %s\n", path);
   err = hwloc_dlforeachfile(path, hwloc__dlforeach_cb, NULL);
   if (err)
     goto out_with_init;
@@ -364,14 +363,14 @@ hwloc_disc_component_register(struct hwloc_disc_component *component,
   /* check that the component name is valid */
   if (!strcmp(component->name, HWLOC_COMPONENT_STOP_NAME)) {
     if (hwloc_components_verbose)
-      fprintf(stderr, "Cannot register discovery component with reserved name `" HWLOC_COMPONENT_STOP_NAME "'\n");
+      fprintf(stderr, "hwloc: Cannot register discovery component with reserved name `" HWLOC_COMPONENT_STOP_NAME "'\n");
     return -1;
   }
   if (strchr(component->name, HWLOC_COMPONENT_EXCLUDE_CHAR)
       || strchr(component->name, HWLOC_COMPONENT_PHASESEP_CHAR)
       || strcspn(component->name, HWLOC_COMPONENT_SEPS) != strlen(component->name)) {
     if (hwloc_components_verbose)
-      fprintf(stderr, "Cannot register discovery component with name `%s' containing reserved characters `%c" HWLOC_COMPONENT_SEPS "'\n",
+      fprintf(stderr, "hwloc: Cannot register discovery component with name `%s' containing reserved characters `%c" HWLOC_COMPONENT_SEPS "'\n",
 	      component->name, HWLOC_COMPONENT_EXCLUDE_CHAR);
     return -1;
   }
@@ -386,8 +385,9 @@ hwloc_disc_component_register(struct hwloc_disc_component *component,
 				   |HWLOC_DISC_PHASE_MISC
 				   |HWLOC_DISC_PHASE_ANNOTATE
 				   |HWLOC_DISC_PHASE_TWEAK))) {
-    fprintf(stderr, "Cannot register discovery component `%s' with invalid phases 0x%x\n",
+    if (HWLOC_SHOW_CRITICAL_ERRORS())
-	    component->name, component->phases);
+      fprintf(stderr, "hwloc: Cannot register discovery component `%s' with invalid phases 0x%x\n",
+              component->name, component->phases);
     return -1;
   }
 
@@ -398,13 +398,13 @@ hwloc_disc_component_register(struct hwloc_disc_component *component,
       if ((*prev)->priority < component->priority) {
 	/* drop the existing component */
 	if (hwloc_components_verbose)
-	  fprintf(stderr, "Dropping previously registered discovery component `%s', priority %u lower than new one %u\n",
+	  fprintf(stderr, "hwloc: Dropping previously registered discovery component `%s', priority %u lower than new one %u\n",
 		  (*prev)->name, (*prev)->priority, component->priority);
 	*prev = (*prev)->next;
       } else {
 	/* drop the new one */
 	if (hwloc_components_verbose)
-	  fprintf(stderr, "Ignoring new discovery component `%s', priority %u lower than previously registered one %u\n",
+	  fprintf(stderr, "hwloc: Ignoring new discovery component `%s', priority %u lower than previously registered one %u\n",
 		  component->name, component->priority, (*prev)->priority);
 	return -1;
       }
@@ -412,7 +412,7 @@ hwloc_disc_component_register(struct hwloc_disc_component *component,
     prev = &((*prev)->next);
   }
   if (hwloc_components_verbose)
-    fprintf(stderr, "Registered discovery component `%s' phases 0x%x with priority %u (%s%s)\n",
+    fprintf(stderr, "hwloc: Registered discovery component `%s' phases 0x%x with priority %u (%s%s)\n",
 	    component->name, component->phases, component->priority,
 	    filename ? "from plugin " : "statically build", filename ? filename : "");
 
@@ -475,15 +475,16 @@ hwloc_components_init(void)
   /* hwloc_static_components is created by configure in static-components.h */
   for(i=0; NULL != hwloc_static_components[i]; i++) {
     if (hwloc_static_components[i]->flags) {
-      fprintf(stderr, "Ignoring static component with invalid flags %lx\n",
+      if (HWLOC_SHOW_CRITICAL_ERRORS())
-	      hwloc_static_components[i]->flags);
+        fprintf(stderr, "hwloc: Ignoring static component with invalid flags %lx\n",
+                hwloc_static_components[i]->flags);
       continue;
     }
 
     /* initialize the component */
     if (hwloc_static_components[i]->init && hwloc_static_components[i]->init(0) < 0) {
       if (hwloc_components_verbose)
-	fprintf(stderr, "Ignoring static component, failed to initialize\n");
+	fprintf(stderr, "hwloc: Ignoring static component, failed to initialize\n");
       continue;
     }
     /* queue ->finalize() callback if any */
@@ -503,15 +504,16 @@ hwloc_components_init(void)
 #ifdef HWLOC_HAVE_PLUGINS
   for(desc = hwloc_plugins; NULL != desc; desc = desc->next) {
     if (desc->component->flags) {
-      fprintf(stderr, "Ignoring plugin `%s' component with invalid flags %lx\n",
+      if (HWLOC_SHOW_CRITICAL_ERRORS())
-	      desc->name, desc->component->flags);
+        fprintf(stderr, "hwloc: Ignoring plugin `%s' component with invalid flags %lx\n",
+                desc->name, desc->component->flags);
       continue;
     }
 
     /* initialize the component */
     if (desc->component->init && desc->component->init(0) < 0) {
       if (hwloc_components_verbose)
-	fprintf(stderr, "Ignoring plugin `%s', failed to initialize\n", desc->name);
+	fprintf(stderr, "hwloc: Ignoring plugin `%s', failed to initialize\n", desc->name);
       continue;
     }
     /* queue ->finalize() callback if any */
@@ -608,7 +610,7 @@ hwloc_disc_component_blacklist_one(struct hwloc_topology *topology,
     /* replace linuxpci and linuxio with linux (with IO phases)
      * for backward compatibility with pre-v2.0 and v2.0 respectively */
     if (hwloc_components_verbose)
-      fprintf(stderr, "Replacing deprecated component `%s' with `linux' IO phases in blacklisting\n", name);
+      fprintf(stderr, "hwloc: Replacing deprecated component `%s' with `linux' IO phases in blacklisting\n", name);
     comp = hwloc_disc_component_find("linux", NULL);
     phases = HWLOC_DISC_PHASE_PCI | HWLOC_DISC_PHASE_IO | HWLOC_DISC_PHASE_MISC | HWLOC_DISC_PHASE_ANNOTATE;
 
@@ -624,7 +626,7 @@ hwloc_disc_component_blacklist_one(struct hwloc_topology *topology,
   }
 
   if (hwloc_components_verbose)
-    fprintf(stderr, "Blacklisting component `%s` phases 0x%x\n", comp->name, phases);
+    fprintf(stderr, "hwloc: Blacklisting component `%s` phases 0x%x\n", comp->name, phases);
 
   for(i=0; i<topology->nr_blacklisted_components; i++) {
     if (topology->blacklisted_components[i].component == comp) {
@@ -727,7 +729,7 @@ hwloc_disc_component_try_enable(struct hwloc_topology *topology,
     if (hwloc_components_verbose)
       /* do not warn if envvar_forced since system-wide HWLOC_COMPONENTS must be silently ignored after set_xml() etc.
        */
-      fprintf(stderr, "Excluding discovery component `%s' phases 0x%x, conflicts with excludes 0x%x\n",
+      fprintf(stderr, "hwloc: Excluding discovery component `%s' phases 0x%x, conflicts with excludes 0x%x\n",
 	      comp->name, comp->phases, topology->backend_excluded_phases);
     return -1;
   }
@@ -735,8 +737,8 @@ hwloc_disc_component_try_enable(struct hwloc_topology *topology,
   backend = comp->instantiate(topology, comp, topology->backend_excluded_phases | blacklisted_phases,
 			      NULL, NULL, NULL);
   if (!backend) {
-    if (hwloc_components_verbose || envvar_forced)
+    if (hwloc_components_verbose || (envvar_forced && HWLOC_SHOW_CRITICAL_ERRORS()))
-      fprintf(stderr, "Failed to instantiate discovery component `%s'\n", comp->name);
+      fprintf(stderr, "hwloc: Failed to instantiate discovery component `%s'\n", comp->name);
     return -1;
   }
 
@@ -817,7 +819,7 @@ hwloc_disc_components_enable_others(struct hwloc_topology *topology)
 	name = curenv;
 	if (!strcmp(name, "linuxpci") || !strcmp(name, "linuxio")) {
 	  if (hwloc_components_verbose)
-	    fprintf(stderr, "Replacing deprecated component `%s' with `linux' in envvar forcing\n", name);
+	    fprintf(stderr, "hwloc: Replacing deprecated component `%s' with `linux' in envvar forcing\n", name);
 	  name = "linux";
 	}
 
@@ -832,7 +834,8 @@ hwloc_disc_components_enable_others(struct hwloc_topology *topology)
 	  if (comp->phases & ~blacklisted_phases)
 	    hwloc_disc_component_try_enable(topology, comp, 1 /* envvar forced */, blacklisted_phases);
 	} else {
-	  fprintf(stderr, "Cannot find discovery component `%s'\n", name);
+          if (HWLOC_SHOW_CRITICAL_ERRORS())
+            fprintf(stderr, "hwloc: Cannot find discovery component `%s'\n", name);
 	}
 
 	/* restore chars (the second loop below needs env to be unmodified) */
@@ -864,7 +867,7 @@ hwloc_disc_components_enable_others(struct hwloc_topology *topology)
 
       if (!(comp->phases & ~blacklisted_phases)) {
 	if (hwloc_components_verbose)
-	  fprintf(stderr, "Excluding blacklisted discovery component `%s' phases 0x%x\n",
+	  fprintf(stderr, "hwloc: Excluding blacklisted discovery component `%s' phases 0x%x\n",
 		  comp->name, comp->phases);
 	goto nextcomp;
       }
@@ -879,7 +882,7 @@ nextcomp:
     /* print a summary */
     int first = 1;
     backend = topology->backends;
-    fprintf(stderr, "Final list of enabled discovery components: ");
+    fprintf(stderr, "hwloc: Final list of enabled discovery components: ");
     while (backend != NULL) {
       fprintf(stderr, "%s%s(0x%x)", first ? "" : ",", backend->component->name, backend->phases);
       backend = backend->next;
@@ -935,7 +938,7 @@ hwloc_backend_alloc(struct hwloc_topology *topology,
   /* filter-out component phases that are excluded */
   backend->phases = component->phases & ~topology->backend_excluded_phases;
   if (backend->phases != component->phases && hwloc_components_verbose)
-    fprintf(stderr, "Trying discovery component `%s' with phases 0x%x instead of 0x%x\n",
+    fprintf(stderr, "hwloc: Trying discovery component `%s' with phases 0x%x instead of 0x%x\n",
 	    component->name, backend->phases, component->phases);
   backend->flags = 0;
   backend->discover = NULL;
@@ -963,8 +966,9 @@ hwloc_backend_enable(struct hwloc_backend *backend)
 
   /* check backend flags */
   if (backend->flags) {
-    fprintf(stderr, "Cannot enable discovery component `%s' phases 0x%x with unknown flags %lx\n",
+    if (HWLOC_SHOW_CRITICAL_ERRORS())
-	    backend->component->name, backend->component->phases, backend->flags);
+      fprintf(stderr, "hwloc: Cannot enable discovery component `%s' phases 0x%x with unknown flags %lx\n",
+              backend->component->name, backend->component->phases, backend->flags);
     return -1;
   }
 
@@ -973,7 +977,7 @@ hwloc_backend_enable(struct hwloc_backend *backend)
   while (NULL != *pprev) {
     if ((*pprev)->component == backend->component) {
       if (hwloc_components_verbose)
-	fprintf(stderr, "Cannot enable  discovery component `%s' phases 0x%x twice\n",
+	fprintf(stderr, "hwloc: Cannot enable  discovery component `%s' phases 0x%x twice\n",
 		backend->component->name, backend->component->phases);
       hwloc_backend_disable(backend);
       errno = EBUSY;
@@ -983,7 +987,7 @@ hwloc_backend_enable(struct hwloc_backend *backend)
   }
 
   if (hwloc_components_verbose)
-    fprintf(stderr, "Enabling discovery component `%s' with phases 0x%x (among 0x%x)\n",
+    fprintf(stderr, "hwloc: Enabling discovery component `%s' with phases 0x%x (among 0x%x)\n",
 	    backend->component->name, backend->phases, backend->component->phases);
 
   /* enqueue at the end */
@@ -1067,7 +1071,7 @@ hwloc_backends_disable_all(struct hwloc_topology *topology)
   while (NULL != (backend = topology->backends)) {
     struct hwloc_backend *next = backend->next;
     if (hwloc_components_verbose)
-      fprintf(stderr, "Disabling discovery component `%s'\n",
+      fprintf(stderr, "hwloc: Disabling discovery component `%s'\n",
 	      backend->component->name);
     hwloc_backend_disable(backend);
     topology->backends = next;

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

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2020 Inria.  All rights reserved.
+ * Copyright © 2020-2024 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -42,11 +42,15 @@ hwloc_internal_cpukinds_dup(hwloc_topology_t new, hwloc_topology_t old)
   struct hwloc_internal_cpukind_s *kinds;
   unsigned i;
 
+  if (!old->nr_cpukinds)
+    return 0;
+
   kinds = hwloc_tma_malloc(tma, old->nr_cpukinds * sizeof(*kinds));
   if (!kinds)
     return -1;
   new->cpukinds = kinds;
   new->nr_cpukinds = old->nr_cpukinds;
+  new->nr_cpukinds_allocated = old->nr_cpukinds;
   memcpy(kinds, old->cpukinds, old->nr_cpukinds * sizeof(*kinds));
 
   for(i=0;i<old->nr_cpukinds; i++) {
@@ -270,7 +274,7 @@ hwloc__cpukinds_check_duplicate_rankings(struct hwloc_topology *topology)
   unsigned i,j;
   for(i=0; i<topology->nr_cpukinds; i++)
     for(j=i+1; j<topology->nr_cpukinds; j++)
-      if (topology->cpukinds[i].forced_efficiency == topology->cpukinds[j].forced_efficiency)
+      if (topology->cpukinds[i].ranking_value == topology->cpukinds[j].ranking_value)
         /* if any duplicate, fail */
         return -1;
   return 0;
@@ -343,7 +347,8 @@ enum hwloc_cpukinds_ranking {
   HWLOC_CPUKINDS_RANKING_DEFAULT, /* forced + frequency on ARM, forced + coretype_frequency otherwise */
   HWLOC_CPUKINDS_RANKING_NO_FORCED_EFFICIENCY, /* default without forced */
   HWLOC_CPUKINDS_RANKING_FORCED_EFFICIENCY,
-  HWLOC_CPUKINDS_RANKING_CORETYPE_FREQUENCY,
+  HWLOC_CPUKINDS_RANKING_CORETYPE_FREQUENCY, /* either coretype or frequency or both */
+  HWLOC_CPUKINDS_RANKING_CORETYPE_FREQUENCY_STRICT, /* both coretype and frequency are required */
   HWLOC_CPUKINDS_RANKING_CORETYPE,
   HWLOC_CPUKINDS_RANKING_FREQUENCY,
   HWLOC_CPUKINDS_RANKING_FREQUENCY_MAX,
@@ -358,9 +363,9 @@ hwloc__cpukinds_try_rank_by_info(struct hwloc_topology *topology,
 {
   unsigned i;
 
-  if (HWLOC_CPUKINDS_RANKING_CORETYPE_FREQUENCY == heuristics) {
+  if (HWLOC_CPUKINDS_RANKING_CORETYPE_FREQUENCY_STRICT == heuristics) {
-    hwloc_debug("Trying to rank cpukinds by coretype+frequency...\n");
+    hwloc_debug("Trying to rank cpukinds by coretype+frequency_strict...\n");
-    /* we need intel_core_type + (base or max freq) for all kinds */
+    /* we need intel_core_type AND (base or max freq) for all kinds */
     if (!summary->have_intel_core_type
         || (!summary->have_max_freq && !summary->have_base_freq))
       return -1;
@@ -373,6 +378,21 @@ hwloc__cpukinds_try_rank_by_info(struct hwloc_topology *topology,
         kind->ranking_value = (summary->summaries[i].intel_core_type << 20) + summary->summaries[i].max_freq;
     }
 
+  } else if (HWLOC_CPUKINDS_RANKING_CORETYPE_FREQUENCY == heuristics) {
+    hwloc_debug("Trying to rank cpukinds by coretype+frequency...\n");
+    /* we need intel_core_type OR (base or max freq) for all kinds */
+    if (!summary->have_intel_core_type
+        && (!summary->have_max_freq && !summary->have_base_freq))
+      return -1;
+    /* rank first by coretype (Core>>Atom) then by frequency, base if available, max otherwise */
+    for(i=0; i<topology->nr_cpukinds; i++) {
+      struct hwloc_internal_cpukind_s *kind = &topology->cpukinds[i];
+      if (summary->have_base_freq)
+        kind->ranking_value = (summary->summaries[i].intel_core_type << 20) + summary->summaries[i].base_freq;
+      else
+        kind->ranking_value = (summary->summaries[i].intel_core_type << 20) + summary->summaries[i].max_freq;
+    }
+
   } else if (HWLOC_CPUKINDS_RANKING_CORETYPE == heuristics) {
     hwloc_debug("Trying to rank cpukinds by coretype...\n");
     /* we need intel_core_type */
@@ -429,7 +449,9 @@ static int hwloc__cpukinds_compare_ranking_values(const void *_a, const void *_b
 {
   const struct hwloc_internal_cpukind_s *a = _a;
   const struct hwloc_internal_cpukind_s *b = _b;
-  return a->ranking_value - b->ranking_value;
+  uint64_t arv = a->ranking_value;
+  uint64_t brv = b->ranking_value;
+  return arv < brv ? -1 : arv > brv ? 1 : 0;
 }
 
 /* this function requires ranking values to be unique */
@@ -469,6 +491,8 @@ hwloc_internal_cpukinds_rank(struct hwloc_topology *topology)
       heuristics = HWLOC_CPUKINDS_RANKING_NONE;
     else if (!strcmp(env, "coretype+frequency"))
       heuristics = HWLOC_CPUKINDS_RANKING_CORETYPE_FREQUENCY;
+    else if (!strcmp(env, "coretype+frequency_strict"))
+      heuristics = HWLOC_CPUKINDS_RANKING_CORETYPE_FREQUENCY_STRICT;
     else if (!strcmp(env, "coretype"))
       heuristics = HWLOC_CPUKINDS_RANKING_CORETYPE;
     else if (!strcmp(env, "frequency"))
@@ -481,16 +505,14 @@ hwloc_internal_cpukinds_rank(struct hwloc_topology *topology)
       heuristics = HWLOC_CPUKINDS_RANKING_FORCED_EFFICIENCY;
     else if (!strcmp(env, "no_forced_efficiency"))
       heuristics = HWLOC_CPUKINDS_RANKING_NO_FORCED_EFFICIENCY;
-    else if (!hwloc_hide_errors())
+    else if (HWLOC_SHOW_CRITICAL_ERRORS())
-      fprintf(stderr, "Failed to recognize HWLOC_CPUKINDS_RANKING value %s\n", env);
+      fprintf(stderr, "hwloc: Failed to recognize HWLOC_CPUKINDS_RANKING value %s\n", env);
   }
 
   if (heuristics == HWLOC_CPUKINDS_RANKING_DEFAULT
       || heuristics == HWLOC_CPUKINDS_RANKING_NO_FORCED_EFFICIENCY) {
     /* default is forced_efficiency first */
     struct hwloc_cpukinds_info_summary summary;
-    enum hwloc_cpukinds_ranking subheuristics;
-    const char *arch;
 
     if (heuristics == HWLOC_CPUKINDS_RANKING_DEFAULT)
       hwloc_debug("Using default ranking strategy...\n");
@@ -508,16 +530,7 @@ hwloc_internal_cpukinds_rank(struct hwloc_topology *topology)
       goto failed;
     hwloc__cpukinds_summarize_info(topology, &summary);
 
-    arch = hwloc_obj_get_info_by_name(topology->levels[0][0], "Architecture");
+    err = hwloc__cpukinds_try_rank_by_info(topology, HWLOC_CPUKINDS_RANKING_CORETYPE_FREQUENCY, &summary);
-    /* TODO: rather coretype_frequency only on x86/Intel? */
-    if (arch && (!strncmp(arch, "arm", 3) || !strncmp(arch, "aarch", 5)))
-      /* then frequency on ARM */
-      subheuristics = HWLOC_CPUKINDS_RANKING_FREQUENCY;
-    else
-      /* or coretype+frequency otherwise */
-      subheuristics = HWLOC_CPUKINDS_RANKING_CORETYPE_FREQUENCY;
-
-    err = hwloc__cpukinds_try_rank_by_info(topology, subheuristics, &summary);
     free(summary.summaries);
     if (!err)
       goto ready;

src/3rdparty/hwloc/src/diff.c

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2013-2020 Inria.  All rights reserved.
+ * Copyright © 2013-2023 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -218,7 +218,7 @@ hwloc_diff_trees(hwloc_topology_t topo1, hwloc_obj_t obj1,
 		struct hwloc_info_s *info1 = &obj1->infos[i], *info2 = &obj2->infos[i];
 		if (strcmp(info1->name, info2->name))
 			goto out_too_complex;
-		if (strcmp(obj1->infos[i].value, obj2->infos[i].value)) {
+		if (strcmp(info1->value, info2->value)) {
 			err = hwloc_append_diff_obj_attr_string(obj1,
 								HWLOC_TOPOLOGY_DIFF_OBJ_ATTR_INFO,
 								info1->name,
@@ -411,6 +411,30 @@ int hwloc_topology_diff_build(hwloc_topology_t topo1,
           }
         }
 
+        if (!err) {
+          /* cpukinds */
+          if (topo1->nr_cpukinds != topo2->nr_cpukinds)
+            goto roottoocomplex;
+          for(i=0; i<topo1->nr_cpukinds; i++) {
+            struct hwloc_internal_cpukind_s *ic1 = &topo1->cpukinds[i];
+            struct hwloc_internal_cpukind_s *ic2 = &topo2->cpukinds[i];
+            unsigned j;
+            if (!hwloc_bitmap_isequal(ic1->cpuset, ic2->cpuset)
+                || ic1->efficiency != ic2->efficiency
+                || ic1->forced_efficiency != ic2->forced_efficiency
+                || ic1->ranking_value != ic2->ranking_value
+                || ic1->nr_infos != ic2->nr_infos)
+              goto roottoocomplex;
+            for(j=0; j<ic1->nr_infos; j++) {
+              struct hwloc_info_s *info1 = &ic1->infos[j], *info2 = &ic2->infos[j];
+              if (strcmp(info1->name, info2->name)
+                  || strcmp(info1->value, info2->value)) {
+                goto roottoocomplex;
+              }
+            }
+          }
+        }
+
 	return err;
 
  roottoocomplex:

src/3rdparty/hwloc/src/distances.c

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2010-2020 Inria.  All rights reserved.
+ * Copyright © 2010-2025 Inria.  All rights reserved.
  * Copyright © 2011-2012 Université Bordeaux
  * Copyright © 2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -17,6 +17,37 @@
 static struct hwloc_internal_distances_s *
 hwloc__internal_distances_from_public(hwloc_topology_t topology, struct hwloc_distances_s *distances);
 
+static void
+hwloc__groups_by_distances(struct hwloc_topology *topology, unsigned nbobjs, struct hwloc_obj **objs, uint64_t *values, unsigned long kind, unsigned nbaccuracies, float *accuracies, int needcheck);
+
+static void
+hwloc_internal_distances_restrict(hwloc_obj_t *objs,
+				  uint64_t *indexes,
+				  hwloc_obj_type_t *different_types,
+				  uint64_t *values,
+				  unsigned nbobjs, unsigned disappeared);
+
+static void
+hwloc_internal_distances_print_matrix(struct hwloc_internal_distances_s *dist)
+{
+  unsigned nbobjs = dist->nbobjs;
+  hwloc_obj_t *objs = dist->objs;
+  hwloc_uint64_t *values = dist->values;
+  int gp = !HWLOC_DIST_TYPE_USE_OS_INDEX(dist->unique_type);
+  unsigned i, j;
+
+  fprintf(stderr, "%s", gp ? "gp_index" : "os_index");
+  for(j=0; j<nbobjs; j++)
+    fprintf(stderr, " % 5d", (int)(gp ? objs[j]->gp_index : objs[j]->os_index));
+  fprintf(stderr, "\n");
+  for(i=0; i<nbobjs; i++) {
+    fprintf(stderr, "  % 5d", (int)(gp ? objs[i]->gp_index : objs[i]->os_index));
+    for(j=0; j<nbobjs; j++)
+      fprintf(stderr, " % 5lld", (long long) values[i*nbobjs + j]);
+    fprintf(stderr, "\n");
+  }
+}
+
 /******************************************************
  * Global init, prepare, destroy, dup
  */
@@ -244,27 +275,33 @@ int hwloc_distances_release_remove(hwloc_topology_t topology,
   return 0;
 }
 
-/******************************************************
+/*********************************************************
- * Add distances to the topology
+ * Backend functions for adding distances to the topology
  */
 
+/* cancel a distances handle. only needed internally for now */
 static void
-hwloc__groups_by_distances(struct hwloc_topology *topology, unsigned nbobjs, struct hwloc_obj **objs, uint64_t *values, unsigned long kind, unsigned nbaccuracies, float *accuracies, int needcheck);
+hwloc_backend_distances_add__cancel(struct hwloc_internal_distances_s *dist)
+{
+  /* everything is set to NULL in hwloc_backend_distances_add_create() */
+  free(dist->name);
+  free(dist->indexes);
+  free(dist->objs);
+  free(dist->different_types);
+  free(dist->values);
+  free(dist);
+}
 
-/* insert a distance matrix in the topology.
+/* prepare a distances handle for later commit in the topology.
- * the caller gives us the distances and objs pointers, we'll free them later.
+ * we duplicate the caller's name.
  */
-static int
+hwloc_backend_distances_add_handle_t
-hwloc_internal_distances__add(hwloc_topology_t topology, const char *name,
+hwloc_backend_distances_add_create(hwloc_topology_t topology,
-			      hwloc_obj_type_t unique_type, hwloc_obj_type_t *different_types,
+                                   const char *name, unsigned long kind, unsigned long flags)
-			      unsigned nbobjs, hwloc_obj_t *objs, uint64_t *indexes, uint64_t *values,
-			      unsigned long kind, unsigned iflags)
 {
   struct hwloc_internal_distances_s *dist;
 
-  if (different_types) {
+  if (flags) {
-    kind |= HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES; /* the user isn't forced to give it */
-  } else if (kind & HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES) {
     errno = EINVAL;
     goto err;
   }
@@ -273,110 +310,54 @@ hwloc_internal_distances__add(hwloc_topology_t topology, const char *name,
   if (!dist)
     goto err;
 
-  if (name)
+  if (name) {
     dist->name = strdup(name); /* ignore failure */
-
+    if (!dist->name)
-  dist->unique_type = unique_type;
-  dist->different_types = different_types;
-  dist->nbobjs = nbobjs;
-  dist->kind = kind;
-  dist->iflags = iflags;
-
-  assert(!!(iflags & HWLOC_INTERNAL_DIST_FLAG_OBJS_VALID) == !!objs);
-
-  if (!objs) {
-    assert(indexes);
-    /* we only have indexes, we'll refresh objs from there */
-    dist->indexes = indexes;
-    dist->objs = calloc(nbobjs, sizeof(hwloc_obj_t));
-    if (!dist->objs)
       goto err_with_dist;
-
-  } else {
-    unsigned i;
-    assert(!indexes);
-    /* we only have objs, generate the indexes arrays so that we can refresh objs later */
-    dist->objs = objs;
-    dist->indexes = malloc(nbobjs * sizeof(*dist->indexes));
-    if (!dist->indexes)
-      goto err_with_dist;
-    if (HWLOC_DIST_TYPE_USE_OS_INDEX(dist->unique_type)) {
-      for(i=0; i<nbobjs; i++)
-	dist->indexes[i] = objs[i]->os_index;
-    } else {
-      for(i=0; i<nbobjs; i++)
-	dist->indexes[i] = objs[i]->gp_index;
-    }
   }
 
-  dist->values = values;
+  dist->kind = kind;
+  dist->iflags = HWLOC_INTERNAL_DIST_FLAG_NOT_COMMITTED;
+
+  dist->unique_type = HWLOC_OBJ_TYPE_NONE;
+  dist->different_types = NULL;
+  dist->nbobjs = 0;
+  dist->indexes = NULL;
+  dist->objs = NULL;
+  dist->values = NULL;
 
   dist->id = topology->next_dist_id++;
-
+  return dist;
-  if (topology->last_dist)
-    topology->last_dist->next = dist;
-  else
-    topology->first_dist = dist;
-  dist->prev = topology->last_dist;
-  dist->next = NULL;
-  topology->last_dist = dist;
-  return 0;
 
  err_with_dist:
-  if (name)
+  hwloc_backend_distances_add__cancel(dist);
-    free(dist->name);
-  free(dist);
  err:
-  free(different_types);
+  return NULL;
-  free(objs);
-  free(indexes);
-  free(values);
-  return -1;
 }
 
-int hwloc_internal_distances_add_by_index(hwloc_topology_t topology, const char *name,
+/* attach objects and values to a distances handle.
-					  hwloc_obj_type_t unique_type, hwloc_obj_type_t *different_types, unsigned nbobjs, uint64_t *indexes, uint64_t *values,
+ * on success, objs and values arrays are attached and will be freed with the distances.
-					  unsigned long kind, unsigned long flags)
+ * on failure, the handle is freed.
+ */
+int
+hwloc_backend_distances_add_values(hwloc_topology_t topology __hwloc_attribute_unused,
+                                   hwloc_backend_distances_add_handle_t handle,
+                                   unsigned nbobjs, hwloc_obj_t *objs,
+                                   hwloc_uint64_t *values,
+                                   unsigned long flags)
 {
-  unsigned iflags = 0; /* objs not valid */
+  struct hwloc_internal_distances_s *dist = handle;
-
+  hwloc_obj_type_t unique_type, *different_types = NULL;
-  if (nbobjs < 2) {
+  hwloc_uint64_t *indexes = NULL;
-    errno = EINVAL;
-    goto err;
-  }
-
-  /* cannot group without objects,
-   * and we don't group from XML anyway since the hwloc that generated the XML should have grouped already.
-   */
-  if (flags & HWLOC_DISTANCES_ADD_FLAG_GROUP) {
-    errno = EINVAL;
-    goto err;
-  }
-
-  return hwloc_internal_distances__add(topology, name, unique_type, different_types, nbobjs, NULL, indexes, values, kind, iflags);
-
- err:
-  free(indexes);
-  free(values);
-  free(different_types);
-  return -1;
-}
-
-static void
-hwloc_internal_distances_restrict(hwloc_obj_t *objs,
-				  uint64_t *indexes,
-				  uint64_t *values,
-				  unsigned nbobjs, unsigned disappeared);
-
-int hwloc_internal_distances_add(hwloc_topology_t topology, const char *name,
-				 unsigned nbobjs, hwloc_obj_t *objs, uint64_t *values,
-				 unsigned long kind, unsigned long flags)
-{
-  hwloc_obj_type_t unique_type, *different_types;
   unsigned i, disappeared = 0;
-  unsigned iflags = HWLOC_INTERNAL_DIST_FLAG_OBJS_VALID;
 
-  if (nbobjs < 2) {
+  if (dist->nbobjs || !(dist->iflags & HWLOC_INTERNAL_DIST_FLAG_NOT_COMMITTED)) {
+    /* target distances is already set */
+    errno = EINVAL;
+    goto err;
+  }
+
+  if (flags || nbobjs < 2 || !objs || !values) {
     errno = EINVAL;
     goto err;
   }
@@ -389,15 +370,18 @@ int hwloc_internal_distances_add(hwloc_topology_t topology, const char *name,
     /* some objects are NULL */
     if (disappeared == nbobjs) {
       /* nothing left, drop the matrix */
-      free(objs);
+      errno = ENOENT;
-      free(values);
+      goto err;
-      return 0;
     }
     /* restrict the matrix */
-    hwloc_internal_distances_restrict(objs, NULL, values, nbobjs, disappeared);
+    hwloc_internal_distances_restrict(objs, NULL, NULL, values, nbobjs, disappeared);
     nbobjs -= disappeared;
   }
 
+  indexes = malloc(nbobjs * sizeof(*indexes));
+  if (!indexes)
+    goto err;
+
   unique_type = objs[0]->type;
   for(i=1; i<nbobjs; i++)
     if (objs[i]->type != unique_type) {
@@ -408,16 +392,108 @@ int hwloc_internal_distances_add(hwloc_topology_t topology, const char *name,
     /* heterogeneous types */
     different_types = malloc(nbobjs * sizeof(*different_types));
     if (!different_types)
-      goto err;
+      goto err_with_indexes;
     for(i=0; i<nbobjs; i++)
       different_types[i] = objs[i]->type;
-
-  } else {
-    /* homogeneous types */
-    different_types = NULL;
   }
 
-  if (topology->grouping && (flags & HWLOC_DISTANCES_ADD_FLAG_GROUP) && !different_types) {
+  dist->nbobjs = nbobjs;
+  dist->objs = objs;
+  dist->iflags |= HWLOC_INTERNAL_DIST_FLAG_OBJS_VALID;
+  dist->indexes = indexes;
+  dist->unique_type = unique_type;
+  dist->different_types = different_types;
+  dist->values = values;
+
+  if (different_types)
+    dist->kind |= HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES;
+
+  if (HWLOC_DIST_TYPE_USE_OS_INDEX(dist->unique_type)) {
+      for(i=0; i<nbobjs; i++)
+	dist->indexes[i] = objs[i]->os_index;
+    } else {
+      for(i=0; i<nbobjs; i++)
+	dist->indexes[i] = objs[i]->gp_index;
+    }
+
+  return 0;
+
+ err_with_indexes:
+  free(indexes);
+ err:
+  hwloc_backend_distances_add__cancel(dist);
+  return -1;
+}
+
+/* attach objects and values to a distance handle.
+ * on success, objs and values arrays are attached and will be freed with the distances.
+ * on failure, the handle is freed.
+ */
+static int
+hwloc_backend_distances_add_values_by_index(hwloc_topology_t topology __hwloc_attribute_unused,
+                                            hwloc_backend_distances_add_handle_t handle,
+                                            unsigned nbobjs, hwloc_obj_type_t unique_type, hwloc_obj_type_t *different_types, hwloc_uint64_t *indexes,
+                                            hwloc_uint64_t *values)
+{
+  struct hwloc_internal_distances_s *dist = handle;
+  hwloc_obj_t *objs;
+
+  if (dist->nbobjs || !(dist->iflags & HWLOC_INTERNAL_DIST_FLAG_NOT_COMMITTED)) {
+    /* target distances is already set */
+    errno = EINVAL;
+    goto err;
+  }
+  if (nbobjs < 2 || !indexes || !values || (unique_type == HWLOC_OBJ_TYPE_NONE && !different_types)) {
+    errno = EINVAL;
+    goto err;
+  }
+
+  objs = malloc(nbobjs * sizeof(*objs));
+  if (!objs)
+    goto err;
+
+  dist->nbobjs = nbobjs;
+  dist->objs = objs;
+  dist->indexes = indexes;
+  dist->unique_type = unique_type;
+  dist->different_types = different_types;
+  dist->values = values;
+
+  if (different_types)
+    dist->kind |= HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES;
+
+  return 0;
+
+ err:
+  hwloc_backend_distances_add__cancel(dist);
+  return -1;
+}
+
+/* commit a distances handle.
+ * on failure, the handle is freed with its objects and values arrays.
+ */
+int
+hwloc_backend_distances_add_commit(hwloc_topology_t topology,
+                                   hwloc_backend_distances_add_handle_t handle,
+                                   unsigned long flags)
+{
+  struct hwloc_internal_distances_s *dist = handle;
+
+  if (!dist->nbobjs || !(dist->iflags & HWLOC_INTERNAL_DIST_FLAG_NOT_COMMITTED)) {
+    /* target distances not ready for commit */
+    errno = EINVAL;
+    goto err;
+  }
+
+  if ((flags & HWLOC_DISTANCES_ADD_FLAG_GROUP) && !dist->objs) {
+    /* cannot group without objects,
+     * and we don't group from XML anyway since the hwloc that generated the XML should have grouped already.
+     */
+    errno = EINVAL;
+    goto err;
+  }
+
+  if (topology->grouping && (flags & HWLOC_DISTANCES_ADD_FLAG_GROUP) && !dist->different_types) {
     float full_accuracy = 0.f;
     float *accuracies;
     unsigned nbaccuracies;
@@ -431,26 +507,94 @@ int hwloc_internal_distances_add(hwloc_topology_t topology, const char *name,
     }
 
     if (topology->grouping_verbose) {
-      unsigned j;
-      int gp = !HWLOC_DIST_TYPE_USE_OS_INDEX(unique_type);
       fprintf(stderr, "Trying to group objects using distance matrix:\n");
-      fprintf(stderr, "%s", gp ? "gp_index" : "os_index");
+      hwloc_internal_distances_print_matrix(dist);
-      for(j=0; j<nbobjs; j++)
-	fprintf(stderr, " % 5d", (int)(gp ? objs[j]->gp_index : objs[j]->os_index));
-      fprintf(stderr, "\n");
-      for(i=0; i<nbobjs; i++) {
-	fprintf(stderr, "  % 5d", (int)(gp ? objs[i]->gp_index : objs[i]->os_index));
-	for(j=0; j<nbobjs; j++)
-	  fprintf(stderr, " % 5lld", (long long) values[i*nbobjs + j]);
-	fprintf(stderr, "\n");
-      }
     }
 
-    hwloc__groups_by_distances(topology, nbobjs, objs, values,
+    hwloc__groups_by_distances(topology, dist->nbobjs, dist->objs, dist->values,
-			       kind, nbaccuracies, accuracies, 1 /* check the first matrice */);
+			       dist->kind, nbaccuracies, accuracies, 1 /* check the first matrix */);
   }
 
-  return hwloc_internal_distances__add(topology, name, unique_type, different_types, nbobjs, objs, NULL, values, kind, iflags);
+  if (topology->last_dist)
+    topology->last_dist->next = dist;
+  else
+    topology->first_dist = dist;
+  dist->prev = topology->last_dist;
+  dist->next = NULL;
+  topology->last_dist = dist;
+
+  dist->iflags &= ~HWLOC_INTERNAL_DIST_FLAG_NOT_COMMITTED;
+  return 0;
+
+ err:
+  hwloc_backend_distances_add__cancel(dist);
+  return -1;
+}
+
+/* all-in-one backend function not exported to plugins, only used by XML for now */
+int hwloc_internal_distances_add_by_index(hwloc_topology_t topology, const char *name,
+                                          hwloc_obj_type_t unique_type, hwloc_obj_type_t *different_types, unsigned nbobjs, uint64_t *indexes, uint64_t *values,
+                                          unsigned long kind, unsigned long flags)
+{
+  hwloc_backend_distances_add_handle_t handle;
+  int err;
+
+  handle = hwloc_backend_distances_add_create(topology, name, kind, 0);
+  if (!handle)
+    goto err;
+
+  err = hwloc_backend_distances_add_values_by_index(topology, handle,
+                                                    nbobjs, unique_type, different_types, indexes,
+                                                    values);
+  if (err < 0)
+    goto err;
+
+  /* arrays are now attached to the handle */
+  indexes = NULL;
+  different_types = NULL;
+  values = NULL;
+
+  err = hwloc_backend_distances_add_commit(topology, handle, flags);
+  if (err < 0)
+    goto err;
+
+  return 0;
+
+ err:
+  free(indexes);
+  free(different_types);
+  free(values);
+  return -1;
+}
+
+/* all-in-one backend function not exported to plugins, used by OS backends */
+int hwloc_internal_distances_add(hwloc_topology_t topology, const char *name,
+                                 unsigned nbobjs, hwloc_obj_t *objs, uint64_t *values,
+                                 unsigned long kind, unsigned long flags)
+{
+  hwloc_backend_distances_add_handle_t handle;
+  int err;
+
+  handle = hwloc_backend_distances_add_create(topology, name, kind, 0);
+  if (!handle)
+    goto err;
+
+  err = hwloc_backend_distances_add_values(topology, handle,
+                                           nbobjs, objs,
+                                           values,
+                                           0);
+  if (err < 0)
+    goto err;
+
+  /* arrays are now attached to the handle */
+  objs = NULL;
+  values = NULL;
+
+  err = hwloc_backend_distances_add_commit(topology, handle, flags);
+  if (err < 0)
+    goto err;
+
+  return 0;
 
  err:
   free(objs);
@@ -458,44 +602,54 @@ int hwloc_internal_distances_add(hwloc_topology_t topology, const char *name,
   return -1;
 }
 
+/********************************
+ * User API for adding distances
+ */
+
 #define HWLOC_DISTANCES_KIND_FROM_ALL (HWLOC_DISTANCES_KIND_FROM_OS|HWLOC_DISTANCES_KIND_FROM_USER)
 #define HWLOC_DISTANCES_KIND_MEANS_ALL (HWLOC_DISTANCES_KIND_MEANS_LATENCY|HWLOC_DISTANCES_KIND_MEANS_BANDWIDTH)
-#define HWLOC_DISTANCES_KIND_ALL (HWLOC_DISTANCES_KIND_FROM_ALL|HWLOC_DISTANCES_KIND_MEANS_ALL)
+#define HWLOC_DISTANCES_KIND_ALL (HWLOC_DISTANCES_KIND_FROM_ALL|HWLOC_DISTANCES_KIND_MEANS_ALL|HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES)
 #define HWLOC_DISTANCES_ADD_FLAG_ALL (HWLOC_DISTANCES_ADD_FLAG_GROUP|HWLOC_DISTANCES_ADD_FLAG_GROUP_INACCURATE)
 
-/* The actual function exported to the user
+void * hwloc_distances_add_create(hwloc_topology_t topology,
- */
+                                  const char *name, unsigned long kind,
-int hwloc_distances_add(hwloc_topology_t topology,
+                                  unsigned long flags)
-			unsigned nbobjs, hwloc_obj_t *objs, hwloc_uint64_t *values,
+{
-			unsigned long kind, unsigned long flags)
+  if (!topology->is_loaded) {
+    errno = EINVAL;
+    return NULL;
+  }
+  if (topology->adopted_shmem_addr) {
+    errno = EPERM;
+    return NULL;
+  }
+  if ((kind & ~HWLOC_DISTANCES_KIND_ALL)
+      || hwloc_weight_long(kind & HWLOC_DISTANCES_KIND_FROM_ALL) > 1
+      || hwloc_weight_long(kind & HWLOC_DISTANCES_KIND_MEANS_ALL) > 1) {
+    errno = EINVAL;
+    return NULL;
+  }
+
+  return hwloc_backend_distances_add_create(topology, name, kind, flags);
+}
+
+int hwloc_distances_add_values(hwloc_topology_t topology,
+                               void *handle,
+                               unsigned nbobjs, hwloc_obj_t *objs,
+                               hwloc_uint64_t *values,
+                               unsigned long flags)
 {
   unsigned i;
   uint64_t *_values;
   hwloc_obj_t *_objs;
   int err;
 
-  if (nbobjs < 2 || !objs || !values || !topology->is_loaded) {
-    errno = EINVAL;
-    return -1;
-  }
-  if (topology->adopted_shmem_addr) {
-    errno = EPERM;
-    return -1;
-  }
-  if ((kind & ~HWLOC_DISTANCES_KIND_ALL)
-      || hwloc_weight_long(kind & HWLOC_DISTANCES_KIND_FROM_ALL) != 1
-      || hwloc_weight_long(kind & HWLOC_DISTANCES_KIND_MEANS_ALL) != 1
-      || (flags & ~HWLOC_DISTANCES_ADD_FLAG_ALL)) {
-    errno = EINVAL;
-    return -1;
-  }
-
   /* no strict need to check for duplicates, things shouldn't break */
 
   for(i=1; i<nbobjs; i++)
     if (!objs[i]) {
       errno = EINVAL;
-      return -1;
+      goto out;
     }
 
   /* copy the input arrays and give them to the topology */
@@ -506,22 +660,78 @@ int hwloc_distances_add(hwloc_topology_t topology,
 
   memcpy(_objs, objs, nbobjs*sizeof(hwloc_obj_t));
   memcpy(_values, values, nbobjs*nbobjs*sizeof(*_values));
-  err = hwloc_internal_distances_add(topology, NULL, nbobjs, _objs, _values, kind, flags);
-  if (err < 0)
-    goto out; /* _objs and _values freed in hwloc_internal_distances_add() */
 
-  /* in case we added some groups, see if we need to reconnect */
+  err = hwloc_backend_distances_add_values(topology, handle, nbobjs, _objs, _values, flags);
-  hwloc_topology_reconnect(topology, 0);
+  if (err < 0) {
+    /* handle was canceled inside hwloc_backend_distances_add_values */
+    handle = NULL;
+    goto out_with_arrays;
+  }
 
   return 0;
 
  out_with_arrays:
-  free(_values);
   free(_objs);
+  free(_values);
  out:
+  if (handle)
+    hwloc_backend_distances_add__cancel(handle);
   return -1;
 }
 
+int
+hwloc_distances_add_commit(hwloc_topology_t topology,
+                           void *handle,
+                           unsigned long flags)
+{
+  int err;
+
+  if (flags & ~HWLOC_DISTANCES_ADD_FLAG_ALL) {
+    errno = EINVAL;
+    goto out;
+  }
+
+  err = hwloc_backend_distances_add_commit(topology, handle, flags);
+  if (err < 0) {
+    /* handle was canceled inside hwloc_backend_distances_add_commit */
+    handle = NULL;
+    goto out;
+  }
+
+  /* in case we added some groups, see if we need to reconnect */
+  hwloc__reconnect(topology, 0);
+
+  return 0;
+
+ out:
+  if (handle)
+    hwloc_backend_distances_add__cancel(handle);
+  return -1;
+}
+
+/* deprecated all-in-one user function */
+int hwloc_distances_add(hwloc_topology_t topology,
+			unsigned nbobjs, hwloc_obj_t *objs, hwloc_uint64_t *values,
+			unsigned long kind, unsigned long flags)
+{
+  void *handle;
+  int err;
+
+  handle = hwloc_distances_add_create(topology, NULL, kind, 0);
+  if (!handle)
+    return -1;
+
+  err = hwloc_distances_add_values(topology, handle, nbobjs, objs, values, 0);
+  if (err < 0)
+    return -1;
+
+  err = hwloc_distances_add_commit(topology, handle, flags);
+  if (err < 0)
+    return -1;
+
+  return 0;
+}
+
 /******************************************************
  * Refresh objects in distances
  */
@@ -529,6 +739,7 @@ int hwloc_distances_add(hwloc_topology_t topology,
 static void
 hwloc_internal_distances_restrict(hwloc_obj_t *objs,
 				  uint64_t *indexes,
+                                  hwloc_obj_type_t *different_types,
 				  uint64_t *values,
 				  unsigned nbobjs, unsigned disappeared)
 {
@@ -550,6 +761,8 @@ hwloc_internal_distances_restrict(hwloc_obj_t *objs,
       objs[newi] = objs[i];
       if (indexes)
 	indexes[newi] = indexes[i];
+      if (different_types)
+        different_types[newi] = different_types[i];
       newi++;
     }
 }
@@ -594,7 +807,7 @@ hwloc_internal_distances_refresh_one(hwloc_topology_t topology,
     return -1;
 
   if (disappeared) {
-    hwloc_internal_distances_restrict(objs, dist->indexes, dist->values, nbobjs, disappeared);
+    hwloc_internal_distances_restrict(objs, dist->indexes, dist->different_types, dist->values, nbobjs, disappeared);
     dist->nbobjs -= disappeared;
   }
 
@@ -647,7 +860,7 @@ struct hwloc_distances_container_s {
   struct hwloc_distances_s distances;
 };
 
-#define HWLOC_DISTANCES_CONTAINER_OFFSET ((char*)&((struct hwloc_distances_container_s*)NULL)->distances - (char*)NULL)
+#define HWLOC_DISTANCES_CONTAINER_OFFSET ((uintptr_t)(&((struct hwloc_distances_container_s*)NULL)->distances) - (uintptr_t)NULL)
 #define HWLOC_DISTANCES_CONTAINER(_d) (struct hwloc_distances_container_s *) ( ((char*)_d) - HWLOC_DISTANCES_CONTAINER_OFFSET )
 
 static struct hwloc_internal_distances_s *
@@ -1087,3 +1300,196 @@ hwloc__groups_by_distances(struct hwloc_topology *topology,
  out_with_groupids:
   free(groupids);
 }
+
+static int
+hwloc__distances_transform_remove_null(struct hwloc_distances_s *distances)
+{
+  hwloc_uint64_t *values = distances->values;
+  hwloc_obj_t *objs = distances->objs;
+  unsigned i, nb, nbobjs = distances->nbobjs;
+  hwloc_obj_type_t unique_type;
+
+  for(i=0, nb=0; i<nbobjs; i++)
+    if (objs[i])
+      nb++;
+
+  if (nb < 2) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  if (nb == nbobjs)
+    return 0;
+
+  hwloc_internal_distances_restrict(objs, NULL, NULL, values, nbobjs, nbobjs-nb);
+  distances->nbobjs = nb;
+
+  /* update HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES for convenience */
+  unique_type = objs[0]->type;
+  for(i=1; i<nb; i++)
+    if (objs[i]->type != unique_type) {
+      unique_type = HWLOC_OBJ_TYPE_NONE;
+      break;
+    }
+  if (unique_type == HWLOC_OBJ_TYPE_NONE)
+    distances->kind |= HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES;
+  else
+    distances->kind &= ~HWLOC_DISTANCES_KIND_HETEROGENEOUS_TYPES;
+
+  return 0;
+}
+
+static int
+hwloc__distances_transform_links(struct hwloc_distances_s *distances)
+{
+  /* FIXME: we should look for the greatest common denominator
+   * but we just use the smallest positive value, that's enough for current use-cases.
+   * We'll return -1 in other cases.
+   */
+  hwloc_uint64_t divider, *values = distances->values;
+  unsigned i, nbobjs = distances->nbobjs;
+
+  if (!(distances->kind & HWLOC_DISTANCES_KIND_MEANS_BANDWIDTH)) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  for(i=0; i<nbobjs; i++)
+    values[i*nbobjs+i] = 0;
+
+  /* find the smallest positive value */
+  divider = 0;
+  for(i=0; i<nbobjs*nbobjs; i++)
+    if (values[i] && (!divider || values[i] < divider))
+      divider = values[i];
+
+  if (!divider)
+    /* only zeroes? do nothing */
+    return 0;
+
+  /* check it divides all values */
+  for(i=0; i<nbobjs*nbobjs; i++)
+    if (values[i]%divider) {
+      errno = ENOENT;
+      return -1;
+    }
+
+  /* ok, now divide for real */
+  for(i=0; i<nbobjs*nbobjs; i++)
+    values[i] /= divider;
+
+  return 0;
+}
+
+static __hwloc_inline int is_nvswitch(hwloc_obj_t obj)
+{
+  return obj && obj->subtype && !strcmp(obj->subtype, "NVSwitch");
+}
+
+static int
+hwloc__distances_transform_merge_switch_ports(struct hwloc_distances_s *distances)
+{
+  hwloc_obj_t *objs = distances->objs;
+  hwloc_uint64_t *values = distances->values;
+  unsigned first, i, j, nbobjs = distances->nbobjs;
+
+  /* find the first port */
+  first = (unsigned) -1;
+  for(i=0; i<nbobjs; i++)
+    if (is_nvswitch(objs[i])) {
+      first = i;
+      break;
+    }
+  if (first == (unsigned)-1) {
+    errno = ENOENT;
+    return -1;
+  }
+
+  for(j=i+1; j<nbobjs; j++) {
+    if (is_nvswitch(objs[j])) {
+      /* another port, merge it */
+      unsigned k;
+      for(k=0; k<nbobjs; k++) {
+        if (k==i || k==j)
+          continue;
+        values[k*nbobjs+i] += values[k*nbobjs+j];
+        values[k*nbobjs+j] = 0;
+        values[i*nbobjs+k] += values[j*nbobjs+k];
+        values[j*nbobjs+k] = 0;
+      }
+      values[i*nbobjs+i] += values[j*nbobjs+j];
+      values[j*nbobjs+j] = 0;
+    }
+    /* the caller will also call REMOVE_NULL to remove other ports */
+    objs[j] = NULL;
+  }
+
+  return 0;
+}
+
+static int
+hwloc__distances_transform_transitive_closure(struct hwloc_distances_s *distances)
+{
+  hwloc_obj_t *objs = distances->objs;
+  hwloc_uint64_t *values = distances->values;
+  unsigned nbobjs = distances->nbobjs;
+  unsigned i, j, k;
+
+  for(i=0; i<nbobjs; i++) {
+    hwloc_uint64_t bw_i2sw = 0;
+    if (is_nvswitch(objs[i]))
+      continue;
+    /* count our BW to the switch */
+    for(k=0; k<nbobjs; k++)
+      if (is_nvswitch(objs[k]))
+        bw_i2sw += values[i*nbobjs+k];
+
+    for(j=0; j<nbobjs; j++) {
+      hwloc_uint64_t bw_sw2j = 0;
+      if (i == j || is_nvswitch(objs[j]))
+        continue;
+      /* count our BW from the switch */
+      for(k=0; k<nbobjs; k++)
+        if (is_nvswitch(objs[k]))
+          bw_sw2j += values[k*nbobjs+j];
+
+      /* bandwidth from i to j now gets indirect bandwidth too, min(i2sw,sw2j) */
+      values[i*nbobjs+j] += bw_i2sw > bw_sw2j ? bw_sw2j : bw_i2sw;
+    }
+  }
+
+  return 0;
+}
+
+int
+hwloc_distances_transform(hwloc_topology_t topology __hwloc_attribute_unused,
+                          struct hwloc_distances_s *distances,
+                          enum hwloc_distances_transform_e transform,
+                          void *transform_attr,
+                          unsigned long flags)
+{
+  if (flags || transform_attr) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  switch (transform) {
+  case HWLOC_DISTANCES_TRANSFORM_REMOVE_NULL:
+    return hwloc__distances_transform_remove_null(distances);
+  case HWLOC_DISTANCES_TRANSFORM_LINKS:
+    return hwloc__distances_transform_links(distances);
+  case HWLOC_DISTANCES_TRANSFORM_MERGE_SWITCH_PORTS:
+  {
+    int err;
+    err = hwloc__distances_transform_merge_switch_ports(distances);
+    if (!err)
+      err = hwloc__distances_transform_remove_null(distances);
+    return err;
+  }
+  case HWLOC_DISTANCES_TRANSFORM_TRANSITIVE_CLOSURE:
+    return hwloc__distances_transform_transitive_closure(distances);
+  default:
+    errno = EINVAL;
+    return -1;
+  }
+}

src/3rdparty/hwloc/src/memattrs.c

@@ -1,11 +1,12 @@
 /*
- * Copyright © 2020 Inria.  All rights reserved.
+ * Copyright © 2020-2025 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
 #include "private/autogen/config.h"
 #include "hwloc.h"
 #include "private/private.h"
+#include "private/debug.h"
 
 
 /*****************************
@@ -13,13 +14,26 @@
  */
 
 static __hwloc_inline
-hwloc_uint64_t hwloc__memattr_get_convenience_value(hwloc_memattr_id_t id,
+int hwloc__memattr_get_convenience_value(hwloc_memattr_id_t id,
-                                                    hwloc_obj_t node)
+                                         hwloc_obj_t node,
+                                         hwloc_uint64_t *valuep)
 {
-  if (id == HWLOC_MEMATTR_ID_CAPACITY)
+  if (id == HWLOC_MEMATTR_ID_CAPACITY) {
-    return node->attr->numanode.local_memory;
+    if (node->type != HWLOC_OBJ_NUMANODE) {
-  else if (id == HWLOC_MEMATTR_ID_LOCALITY)
+      errno = EINVAL;
-    return hwloc_bitmap_weight(node->cpuset);
+      return -1;
+    }
+    *valuep = node->attr->numanode.local_memory;
+    return 0;
+  }
+  else if (id == HWLOC_MEMATTR_ID_LOCALITY) {
+    if (!node->cpuset) {
+      errno = EINVAL;
+      return -1;
+    }
+    *valuep = hwloc_bitmap_weight(node->cpuset);
+    return 0;
+  }
   else
     assert(0);
   return 0; /* shut up the compiler */
@@ -49,36 +63,51 @@ hwloc__setup_memattr(struct hwloc_internal_memattr_s *imattr,
 void
 hwloc_internal_memattrs_prepare(struct hwloc_topology *topology)
 {
-#define NR_DEFAULT_MEMATTRS 4
+  topology->memattrs = malloc(HWLOC_MEMATTR_ID_MAX * sizeof(*topology->memattrs));
-  topology->memattrs = malloc(NR_DEFAULT_MEMATTRS * sizeof(*topology->memattrs));
   if (!topology->memattrs)
     return;
 
-  assert(HWLOC_MEMATTR_ID_CAPACITY < NR_DEFAULT_MEMATTRS);
   hwloc__setup_memattr(&topology->memattrs[HWLOC_MEMATTR_ID_CAPACITY],
                        (char *) "Capacity",
                        HWLOC_MEMATTR_FLAG_HIGHER_FIRST,
                        HWLOC_IMATTR_FLAG_STATIC_NAME|HWLOC_IMATTR_FLAG_CONVENIENCE);
 
-  assert(HWLOC_MEMATTR_ID_LOCALITY < NR_DEFAULT_MEMATTRS);
   hwloc__setup_memattr(&topology->memattrs[HWLOC_MEMATTR_ID_LOCALITY],
                        (char *) "Locality",
                        HWLOC_MEMATTR_FLAG_LOWER_FIRST,
                        HWLOC_IMATTR_FLAG_STATIC_NAME|HWLOC_IMATTR_FLAG_CONVENIENCE);
 
-  assert(HWLOC_MEMATTR_ID_BANDWIDTH < NR_DEFAULT_MEMATTRS);
   hwloc__setup_memattr(&topology->memattrs[HWLOC_MEMATTR_ID_BANDWIDTH],
                        (char *) "Bandwidth",
                        HWLOC_MEMATTR_FLAG_HIGHER_FIRST|HWLOC_MEMATTR_FLAG_NEED_INITIATOR,
                        HWLOC_IMATTR_FLAG_STATIC_NAME);
 
-  assert(HWLOC_MEMATTR_ID_LATENCY < NR_DEFAULT_MEMATTRS);
+  hwloc__setup_memattr(&topology->memattrs[HWLOC_MEMATTR_ID_READ_BANDWIDTH],
+                       (char *) "ReadBandwidth",
+                       HWLOC_MEMATTR_FLAG_HIGHER_FIRST|HWLOC_MEMATTR_FLAG_NEED_INITIATOR,
+                       HWLOC_IMATTR_FLAG_STATIC_NAME);
+
+  hwloc__setup_memattr(&topology->memattrs[HWLOC_MEMATTR_ID_WRITE_BANDWIDTH],
+                       (char *) "WriteBandwidth",
+                       HWLOC_MEMATTR_FLAG_HIGHER_FIRST|HWLOC_MEMATTR_FLAG_NEED_INITIATOR,
+                       HWLOC_IMATTR_FLAG_STATIC_NAME);
+
   hwloc__setup_memattr(&topology->memattrs[HWLOC_MEMATTR_ID_LATENCY],
                        (char *) "Latency",
                        HWLOC_MEMATTR_FLAG_LOWER_FIRST|HWLOC_MEMATTR_FLAG_NEED_INITIATOR,
                        HWLOC_IMATTR_FLAG_STATIC_NAME);
 
-  topology->nr_memattrs = NR_DEFAULT_MEMATTRS;
+  hwloc__setup_memattr(&topology->memattrs[HWLOC_MEMATTR_ID_READ_LATENCY],
+                       (char *) "ReadLatency",
+                       HWLOC_MEMATTR_FLAG_LOWER_FIRST|HWLOC_MEMATTR_FLAG_NEED_INITIATOR,
+                       HWLOC_IMATTR_FLAG_STATIC_NAME);
+
+  hwloc__setup_memattr(&topology->memattrs[HWLOC_MEMATTR_ID_WRITE_LATENCY],
+                       (char *) "WriteLatency",
+                       HWLOC_MEMATTR_FLAG_LOWER_FIRST|HWLOC_MEMATTR_FLAG_NEED_INITIATOR,
+                       HWLOC_IMATTR_FLAG_STATIC_NAME);
+
+  topology->nr_memattrs = HWLOC_MEMATTR_ID_MAX;
 }
 
 static void
@@ -127,6 +156,8 @@ hwloc_internal_memattrs_dup(struct hwloc_topology *new, struct hwloc_topology *o
   struct hwloc_internal_memattr_s *imattrs;
   hwloc_memattr_id_t id;
 
+  /* old->nr_memattrs is always > 0 thanks to default memattrs */
+
   imattrs = hwloc_tma_malloc(tma, old->nr_memattrs * sizeof(*imattrs));
   if (!imattrs)
     return -1;
@@ -604,7 +635,7 @@ hwloc_memattr_get_targets(hwloc_topology_t topology,
       if (found<max) {
         targets[found] = node;
         if (values)
-          values[found] = hwloc__memattr_get_convenience_value(id, node);
+          hwloc__memattr_get_convenience_value(id, node, &values[found]);
       }
       found++;
     }
@@ -730,7 +761,7 @@ hwloc_memattr_get_initiators(hwloc_topology_t topology,
   struct hwloc_internal_memattr_target_s *imtg;
   unsigned i, max;
 
-  if (flags) {
+  if (flags || !target_node) {
     errno = EINVAL;
     return -1;
   }
@@ -792,7 +823,7 @@ hwloc_memattr_get_value(hwloc_topology_t topology,
   struct hwloc_internal_memattr_s *imattr;
   struct hwloc_internal_memattr_target_s *imtg;
 
-  if (flags) {
+  if (flags || !target_node) {
     errno = EINVAL;
     return -1;
   }
@@ -805,8 +836,7 @@ hwloc_memattr_get_value(hwloc_topology_t topology,
 
   if (imattr->iflags & HWLOC_IMATTR_FLAG_CONVENIENCE) {
     /* convenience attributes */
-    *valuep = hwloc__memattr_get_convenience_value(id, target_node);
+    return hwloc__memattr_get_convenience_value(id, target_node, valuep);
-    return 0;
   }
 
   /* normal attributes */
@@ -918,7 +948,7 @@ hwloc_memattr_set_value(hwloc_topology_t topology,
 {
   struct hwloc_internal_location_s iloc, *ilocp;
 
-  if (flags) {
+  if (flags || !target_node) {
     errno = EINVAL;
     return -1;
   }
@@ -989,10 +1019,10 @@ hwloc_memattr_get_best_target(hwloc_topology_t topology,
     /* convenience attributes */
     for(j=0; ; j++) {
       hwloc_obj_t node = hwloc_get_obj_by_type(topology, HWLOC_OBJ_NUMANODE, j);
-      hwloc_uint64_t value;
+      hwloc_uint64_t value = 0;
       if (!node)
         break;
-      value = hwloc__memattr_get_convenience_value(id, node);
+      hwloc__memattr_get_convenience_value(id, node, &value);
       hwloc__update_best_target(&best, &best_value, &found,
                                 node, value,
                                 imattr->flags & HWLOC_MEMATTR_FLAG_HIGHER_FIRST);
@@ -1075,7 +1105,7 @@ hwloc_memattr_get_best_initiator(hwloc_topology_t topology,
   int found;
   unsigned i;
 
-  if (flags) {
+  if (flags || !target_node) {
     errno = EINVAL;
     return -1;
   }
@@ -1128,6 +1158,8 @@ match_local_obj_cpuset(hwloc_obj_t node, hwloc_cpuset_t cpuset, unsigned long fl
 {
   if (flags & HWLOC_LOCAL_NUMANODE_FLAG_ALL)
     return 1;
+  if (flags & HWLOC_LOCAL_NUMANODE_FLAG_INTERSECT_LOCALITY)
+    return hwloc_bitmap_intersects(node->cpuset, cpuset);
   if ((flags & HWLOC_LOCAL_NUMANODE_FLAG_LARGER_LOCALITY)
       && hwloc_bitmap_isincluded(cpuset, node->cpuset))
     return 1;
@@ -1150,6 +1182,7 @@ hwloc_get_local_numanode_objs(hwloc_topology_t topology,
 
   if (flags & ~(HWLOC_LOCAL_NUMANODE_FLAG_SMALLER_LOCALITY
                 |HWLOC_LOCAL_NUMANODE_FLAG_LARGER_LOCALITY
+                |HWLOC_LOCAL_NUMANODE_FLAG_INTERSECT_LOCALITY
                 | HWLOC_LOCAL_NUMANODE_FLAG_ALL)) {
     errno = EINVAL;
     return -1;
@@ -1195,3 +1228,750 @@ hwloc_get_local_numanode_objs(hwloc_topology_t topology,
   *nrp = i;
   return 0;
 }
+
+static int compare_nodes_by_os_index(const void *_a, const void *_b)
+{
+  const hwloc_obj_t * a = _a, * b = _b;
+  return (*a)->os_index - (*b)->os_index;
+}
+
+int
+hwloc_topology_get_default_nodeset(hwloc_topology_t topology,
+                                   hwloc_nodeset_t nodeset,
+                                   unsigned long flags)
+{
+  hwloc_obj_t *nodes;
+  hwloc_bitmap_t remainingcpuset;
+  unsigned nrnodes, i;
+  const char *first_subtype;
+
+  if (flags) {
+    errno = EINVAL;
+    goto out;
+  }
+
+  remainingcpuset = hwloc_bitmap_dup(topology->levels[0][0]->cpuset);
+  if (!remainingcpuset)
+    goto out;
+
+  nrnodes = topology->slevels[HWLOC_SLEVEL_NUMANODE].nbobjs;
+  nodes = malloc(nrnodes * sizeof(*nodes));
+  if (!nodes)
+    goto out_with_remainingcpuset;
+
+  memcpy(nodes, topology->slevels[HWLOC_SLEVEL_NUMANODE].objs, nrnodes * sizeof(*nodes));
+  qsort(nodes, nrnodes, sizeof(*nodes), compare_nodes_by_os_index);
+
+  hwloc_bitmap_zero(nodeset);
+
+  /* always take the first node (FIXME: except if unexpected subtype?) */
+  first_subtype = nodes[0]->subtype;
+  hwloc_bitmap_set(nodeset, nodes[0]->os_index);
+  hwloc_bitmap_andnot(remainingcpuset, remainingcpuset, nodes[0]->cpuset);
+
+  /* use all non-intersecting nodes with same subtype */
+  for(i=1; i<nrnodes; i++) {
+    /* check same or no subtype */
+    if (first_subtype) {
+      if (!nodes[i]->subtype || strcmp(first_subtype, nodes[i]->subtype))
+        continue;
+    } else if (nodes[i]->subtype) {
+      continue;
+    }
+    /* take non-overlapping nodes */
+    if (hwloc_bitmap_isincluded(nodes[i]->cpuset, remainingcpuset) /* can be empty */) {
+      hwloc_bitmap_set(nodeset, nodes[i]->os_index);
+      hwloc_bitmap_andnot(remainingcpuset, remainingcpuset, nodes[i]->cpuset);
+    }
+    /* more needed? */
+    if (hwloc_bitmap_iszero(remainingcpuset))
+      goto done;
+  }
+
+  /* find more nodes to cover the entire topology cpuset.
+   * only take what's necessary: first nodes, non-empty */
+  for(i=1; i<nrnodes; i++) {
+    /* already taken? */
+    if (hwloc_bitmap_isset(nodeset, i))
+      continue;
+    /* take non-overlapping nodes, except empty  */
+    if (hwloc_bitmap_isincluded(nodes[i]->cpuset, remainingcpuset)
+        && !hwloc_bitmap_iszero(nodes[i]->cpuset)) {
+      hwloc_bitmap_set(nodeset, nodes[i]->os_index);
+      hwloc_bitmap_andnot(remainingcpuset, remainingcpuset, nodes[i]->cpuset);
+    }
+    /* more needed? */
+    if (hwloc_bitmap_iszero(remainingcpuset))
+      goto done;
+  }
+
+ done:
+  free(nodes);
+  hwloc_bitmap_free(remainingcpuset);
+  return 0;
+
+ out_with_remainingcpuset:
+  hwloc_bitmap_free(remainingcpuset);
+ out:
+  return -1;
+}
+
+
+/**************************************
+ * Using memattrs to identify HBM/DRAM
+ */
+
+enum hwloc_memory_tier_type_e {
+  /* WARNING: keep higher BW types first for compare_tiers_by_bw_and_type() when BW info is missing */
+  HWLOC_MEMORY_TIER_HBM  = 1UL<<0,
+  HWLOC_MEMORY_TIER_DRAM = 1UL<<1,
+  HWLOC_MEMORY_TIER_GPU  = 1UL<<2,
+  HWLOC_MEMORY_TIER_SPM  = 1UL<<3, /* Specific-Purpose Memory is usually HBM, we'll use BW to confirm or force*/
+  HWLOC_MEMORY_TIER_NVM  = 1UL<<4,
+  HWLOC_MEMORY_TIER_CXL  = 1UL<<5
+};
+typedef unsigned long hwloc_memory_tier_type_t;
+#define HWLOC_MEMORY_TIER_UNKNOWN 0UL
+
+static const char * hwloc_memory_tier_type_snprintf(hwloc_memory_tier_type_t type)
+{
+  switch (type) {
+  case HWLOC_MEMORY_TIER_DRAM: return "DRAM";
+  case HWLOC_MEMORY_TIER_HBM: return "HBM";
+  case HWLOC_MEMORY_TIER_GPU: return "GPUMemory";
+  case HWLOC_MEMORY_TIER_SPM: return "SPM";
+  case HWLOC_MEMORY_TIER_NVM: return "NVM";
+  case HWLOC_MEMORY_TIER_CXL:
+  case HWLOC_MEMORY_TIER_CXL|HWLOC_MEMORY_TIER_DRAM: return "CXL-DRAM";
+  case HWLOC_MEMORY_TIER_CXL|HWLOC_MEMORY_TIER_HBM: return "CXL-HBM";
+  case HWLOC_MEMORY_TIER_CXL|HWLOC_MEMORY_TIER_GPU: return "CXL-GPUMemory";
+  case HWLOC_MEMORY_TIER_CXL|HWLOC_MEMORY_TIER_SPM: return "CXL-SPM";
+  case HWLOC_MEMORY_TIER_CXL|HWLOC_MEMORY_TIER_NVM: return "CXL-NVM";
+  default: return NULL;
+  }
+}
+
+static hwloc_memory_tier_type_t hwloc_memory_tier_type_sscanf(const char *name)
+{
+  if (!strcasecmp(name, "DRAM"))
+    return HWLOC_MEMORY_TIER_DRAM;
+  if (!strcasecmp(name, "HBM"))
+    return HWLOC_MEMORY_TIER_HBM;
+  if (!strcasecmp(name, "GPUMemory"))
+    return HWLOC_MEMORY_TIER_GPU;
+  if (!strcasecmp(name, "SPM"))
+    return HWLOC_MEMORY_TIER_SPM;
+  if (!strcasecmp(name, "NVM"))
+    return HWLOC_MEMORY_TIER_NVM;
+  if (!strcasecmp(name, "CXL-DRAM"))
+    return HWLOC_MEMORY_TIER_CXL|HWLOC_MEMORY_TIER_DRAM;
+  if (!strcasecmp(name, "CXL-HBM"))
+    return HWLOC_MEMORY_TIER_CXL|HWLOC_MEMORY_TIER_HBM;
+  if (!strcasecmp(name, "CXL-GPUMemory"))
+    return HWLOC_MEMORY_TIER_CXL|HWLOC_MEMORY_TIER_GPU;
+  if (!strcasecmp(name, "CXL-SPM"))
+    return HWLOC_MEMORY_TIER_CXL|HWLOC_MEMORY_TIER_SPM;
+  if (!strcasecmp(name, "CXL-NVM"))
+    return HWLOC_MEMORY_TIER_CXL|HWLOC_MEMORY_TIER_NVM;
+  return 0;
+}
+
+/* factorized tier, grouping multiple nodes */
+struct hwloc_memory_tier_s {
+  hwloc_nodeset_t nodeset;
+  uint64_t local_bw_min, local_bw_max;
+  uint64_t local_lat_min, local_lat_max;
+  hwloc_memory_tier_type_t type;
+};
+
+/* early tier discovery, one entry per node */
+struct hwloc_memory_node_info_s {
+  hwloc_obj_t node;
+  uint64_t local_bw;
+  uint64_t local_lat;
+  hwloc_memory_tier_type_t type;
+  unsigned rank;
+};
+
+static int compare_node_infos_by_type_and_bw(const void *_a, const void *_b)
+{
+  const struct hwloc_memory_node_info_s *a = _a, *b = _b;
+  /* sort by type of node first */
+  if (a->type != b->type)
+    return a->type - b->type;
+  /* then by bandwidth */
+  if (a->local_bw > b->local_bw)
+    return -1;
+  else if (a->local_bw < b->local_bw)
+    return 1;
+  return 0;
+}
+
+static int compare_tiers_by_bw_and_type(const void *_a, const void *_b)
+{
+  const struct hwloc_memory_tier_s *a = _a, *b = _b;
+  /* sort by (average) BW first */
+  if (a->local_bw_min && b->local_bw_min) {
+    if (a->local_bw_min + a->local_bw_max > b->local_bw_min + b->local_bw_max)
+      return -1;
+    else if (a->local_bw_min + a->local_bw_max < b->local_bw_min + b->local_bw_max)
+      return 1;
+  }
+  /* then by tier type */
+  if (a->type != b->type)
+    return a->type - b->type;
+  return 0;
+}
+
+static struct hwloc_memory_tier_s *
+hwloc__group_memory_tiers(hwloc_topology_t topology,
+                          unsigned *nr_tiers_p)
+{
+  struct hwloc_internal_memattr_s *imattr_bw, *imattr_lat;
+  struct hwloc_memory_node_info_s *nodeinfos;
+  struct hwloc_memory_tier_s *tiers;
+  unsigned nr_tiers;
+  float bw_threshold = 0.1;
+  float lat_threshold = 0.1;
+  const char *env;
+  unsigned i, j, n;
+
+  n = hwloc_get_nbobjs_by_depth(topology, HWLOC_TYPE_DEPTH_NUMANODE);
+  assert(n);
+
+  env = getenv("HWLOC_MEMTIERS_BANDWIDTH_THRESHOLD");
+  if (env)
+    bw_threshold = atof(env);
+
+  env = getenv("HWLOC_MEMTIERS_LATENCY_THRESHOLD");
+  if (env)
+    lat_threshold = atof(env);
+
+  imattr_bw = &topology->memattrs[HWLOC_MEMATTR_ID_BANDWIDTH];
+  imattr_lat = &topology->memattrs[HWLOC_MEMATTR_ID_LATENCY];
+
+  if (!(imattr_bw->iflags & HWLOC_IMATTR_FLAG_CACHE_VALID))
+    hwloc__imattr_refresh(topology, imattr_bw);
+  if (!(imattr_lat->iflags & HWLOC_IMATTR_FLAG_CACHE_VALID))
+    hwloc__imattr_refresh(topology, imattr_lat);
+
+  nodeinfos = malloc(n * sizeof(*nodeinfos));
+  if (!nodeinfos)
+    return NULL;
+
+  for(i=0; i<n; i++) {
+    hwloc_obj_t node;
+    const char *daxtype;
+    struct hwloc_internal_location_s iloc;
+    struct hwloc_internal_memattr_target_s *imtg;
+
+    node = hwloc_get_obj_by_depth(topology, HWLOC_TYPE_DEPTH_NUMANODE, i);
+    assert(node);
+    nodeinfos[i].node = node;
+
+    /* defaults to unknown */
+    nodeinfos[i].type = HWLOC_MEMORY_TIER_UNKNOWN;
+    nodeinfos[i].local_bw = 0;
+    nodeinfos[i].local_lat = 0;
+
+    daxtype = hwloc_obj_get_info_by_name(node, "DAXType");
+    /* mark NVM, SPM and GPU nodes */
+    if (node->subtype && !strcmp(node->subtype, "GPUMemory"))
+      nodeinfos[i].type = HWLOC_MEMORY_TIER_GPU;
+    else if (daxtype && !strcmp(daxtype, "NVM"))
+      nodeinfos[i].type = HWLOC_MEMORY_TIER_NVM;
+    else if (daxtype && !strcmp(daxtype, "SPM"))
+      nodeinfos[i].type = HWLOC_MEMORY_TIER_SPM;
+    /* add CXL flag */
+    if (hwloc_obj_get_info_by_name(node, "CXLDevice") != NULL) {
+      /* CXL is always SPM for now. HBM and DRAM not possible here yet.
+       * Hence remove all but NVM first.
+       */
+      nodeinfos[i].type &= HWLOC_MEMORY_TIER_NVM;
+      nodeinfos[i].type |= HWLOC_MEMORY_TIER_CXL;
+    }
+
+    /* get local bandwidth */
+    imtg = NULL;
+    for(j=0; j<imattr_bw->nr_targets; j++)
+      if (imattr_bw->targets[j].obj == node) {
+        imtg = &imattr_bw->targets[j];
+        break;
+      }
+    if (imtg && !hwloc_bitmap_iszero(node->cpuset)) {
+      struct hwloc_internal_memattr_initiator_s *imi;
+      iloc.type = HWLOC_LOCATION_TYPE_CPUSET;
+      iloc.location.cpuset = node->cpuset;
+      imi = hwloc__memattr_target_get_initiator(imtg, &iloc, 0);
+      if (imi)
+        nodeinfos[i].local_bw = imi->value;
+    }
+    /* get local latency */
+    imtg = NULL;
+    for(j=0; j<imattr_lat->nr_targets; j++)
+      if (imattr_lat->targets[j].obj == node) {
+        imtg = &imattr_lat->targets[j];
+        break;
+      }
+    if (imtg && !hwloc_bitmap_iszero(node->cpuset)) {
+      struct hwloc_internal_memattr_initiator_s *imi;
+      iloc.type = HWLOC_LOCATION_TYPE_CPUSET;
+      iloc.location.cpuset = node->cpuset;
+      imi = hwloc__memattr_target_get_initiator(imtg, &iloc, 0);
+      if (imi)
+        nodeinfos[i].local_lat = imi->value;
+    }
+  }
+
+  /* Sort nodes by tier type and bandwidth.
+   *
+   * We could also use the existing subtype but it's not clear it'd be better.
+   * For NVIDIA GPU, "GPUMemory" is set in the Linux backend, and used above to set tier type anyway.
+   * For KNL, the Linux backend sets subtypes and memattrs, sorting by memattrs already works fine.
+   * Existing subtypes could have been imported from XML, usually mostly OK except maybe SPM (fallback for I don't know)?
+   * An envvar (or HWLOC_MEMTIERS_REFRESH special value?) could be passed to ignore existing subtypes,
+   * but "GPUMemory" wouldn't be available anymore, we'd have to use something else like "PCIBusId",
+   * but that one might not always be specific to GPU-backed NUMA nodes?
+   */
+  hwloc_debug("Sorting memory node infos...\n");
+  qsort(nodeinfos, n, sizeof(*nodeinfos), compare_node_infos_by_type_and_bw);
+#ifdef HWLOC_DEBUG
+  for(i=0; i<n; i++)
+    hwloc_debug("  node info %u = node L#%u P#%u with info type %lx and local BW %llu lat %llu\n",
+                i,
+                nodeinfos[i].node->logical_index, nodeinfos[i].node->os_index,
+                nodeinfos[i].type,
+                (unsigned long long) nodeinfos[i].local_bw,
+                (unsigned long long) nodeinfos[i].local_lat);
+#endif
+  /* now we have UNKNOWN nodes (sorted by BW only), then known ones */
+
+  /* iterate among them and add a rank value.
+   * start from rank 0 and switch to next rank when the type changes or when the BW or latendy difference is > threshold */
+  hwloc_debug("Starting memory tier #0 and iterating over nodes...\n");
+  nodeinfos[0].rank = 0;
+  for(i=1; i<n; i++) {
+    /* reuse the same rank by default */
+    nodeinfos[i].rank = nodeinfos[i-1].rank;
+    /* comparing type */
+    if (nodeinfos[i].type != nodeinfos[i-1].type) {
+      hwloc_debug("  Switching to memory tier #%u starting with node L#%u P#%u because of type\n",
+                  nodeinfos[i].rank, nodeinfos[i].node->logical_index, nodeinfos[i].node->os_index);
+      nodeinfos[i].rank++;
+      continue;
+    }
+    /* comparing bandwidth */
+    if (nodeinfos[i].local_bw && nodeinfos[i-1].local_bw) {
+      float bw_ratio = (float)nodeinfos[i].local_bw/(float)nodeinfos[i-1].local_bw;
+      if (bw_ratio < 1.)
+        bw_ratio = 1./bw_ratio;
+      if (bw_ratio > 1.0 + bw_threshold) {
+        nodeinfos[i].rank++;
+        hwloc_debug("  Switching to memory tier #%u starting with node L#%u P#%u because of bandwidth\n",
+                    nodeinfos[i].rank, nodeinfos[i].node->logical_index, nodeinfos[i].node->os_index);
+        continue;
+      }
+    }
+    /* comparing latency */
+    if (nodeinfos[i].local_lat && nodeinfos[i-1].local_lat) {
+      float lat_ratio = (float)nodeinfos[i].local_lat/(float)nodeinfos[i-1].local_lat;
+      if (lat_ratio < 1.)
+        lat_ratio = 1./lat_ratio;
+      if (lat_ratio > 1.0 + lat_threshold) {
+        hwloc_debug("  Switching to memory tier #%u starting with node L#%u P#%u because of latency\n",
+                    nodeinfos[i].rank, nodeinfos[i].node->logical_index, nodeinfos[i].node->os_index);
+        nodeinfos[i].rank++;
+        continue;
+      }
+    }
+  }
+  /* FIXME: if there are cpuset-intersecting nodes in same tier, split again? */
+  hwloc_debug("  Found %u tiers total\n", nodeinfos[n-1].rank + 1);
+
+  /* now group nodeinfos into factorized tiers */
+  nr_tiers = nodeinfos[n-1].rank + 1;
+  tiers = calloc(nr_tiers, sizeof(*tiers));
+  if (!tiers)
+    goto out_with_nodeinfos;
+  for(i=0; i<nr_tiers; i++) {
+    tiers[i].nodeset = hwloc_bitmap_alloc();
+    if (!tiers[i].nodeset)
+      goto out_with_tiers;
+    tiers[i].local_bw_min = tiers[i].local_bw_max = 0;
+    tiers[i].local_lat_min = tiers[i].local_lat_max = 0;
+    tiers[i].type = HWLOC_MEMORY_TIER_UNKNOWN;
+  }
+  for(i=0; i<n; i++) {
+    unsigned rank = nodeinfos[i].rank;
+    assert(rank < nr_tiers);
+    hwloc_bitmap_set(tiers[rank].nodeset, nodeinfos[i].node->os_index);
+    assert(tiers[rank].type == HWLOC_MEMORY_TIER_UNKNOWN
+           || tiers[rank].type == nodeinfos[i].type);
+    tiers[rank].type = nodeinfos[i].type;
+    /* nodeinfos are sorted in BW order, no need to compare */
+    if (!tiers[rank].local_bw_min)
+      tiers[rank].local_bw_min = nodeinfos[i].local_bw;
+    tiers[rank].local_bw_max = nodeinfos[i].local_bw;
+    /* compare latencies to update min/max */
+    if (!tiers[rank].local_lat_min || nodeinfos[i].local_lat < tiers[rank].local_lat_min)
+      tiers[rank].local_lat_min = nodeinfos[i].local_lat;
+    if (!tiers[rank].local_lat_max || nodeinfos[i].local_lat > tiers[rank].local_lat_max)
+      tiers[rank].local_lat_max = nodeinfos[i].local_lat;
+  }
+
+  free(nodeinfos);
+  *nr_tiers_p = nr_tiers;
+  return tiers;
+
+ out_with_tiers:
+  for(i=0; i<nr_tiers; i++)
+    hwloc_bitmap_free(tiers[i].nodeset);
+  free(tiers);
+ out_with_nodeinfos:
+  free(nodeinfos);
+  return NULL;
+}
+
+enum hwloc_guess_memtiers_flag {
+  HWLOC_GUESS_MEMTIERS_FLAG_NODE0_IS_DRAM = 1<<0,
+  HWLOC_GUESS_MEMTIERS_FLAG_SPM_IS_HBM = 1<<1
+};
+
+static int
+hwloc__guess_dram_hbm_tiers(struct hwloc_memory_tier_s *tier1,
+                            struct hwloc_memory_tier_s *tier2,
+                            unsigned long flags)
+{
+  struct hwloc_memory_tier_s *tmp;
+
+  if (!tier1->local_bw_min || !tier2->local_bw_min) {
+    hwloc_debug("    Missing BW info\n");
+    return -1;
+  }
+
+  /* reorder tiers by BW */
+  if (tier1->local_bw_min > tier2->local_bw_min) {
+    tmp = tier1; tier1 = tier2; tier2 = tmp;
+  }
+  /* tier1 < tier2 */
+
+  hwloc_debug("    tier1 BW %llu-%llu vs tier2 BW %llu-%llu\n",
+              (unsigned long long) tier1->local_bw_min,
+              (unsigned long long) tier1->local_bw_max,
+              (unsigned long long) tier2->local_bw_min,
+              (unsigned long long) tier2->local_bw_max);
+  if (tier2->local_bw_min <= tier1->local_bw_max * 2) {
+    /* tier2 BW isn't 2x tier1, we cannot guess HBM */
+    hwloc_debug("    BW difference isn't >2x\n");
+    return -1;
+  }
+  /* tier2 BW is >2x tier1 */
+
+  if ((flags & HWLOC_GUESS_MEMTIERS_FLAG_NODE0_IS_DRAM)
+      && hwloc_bitmap_isset(tier2->nodeset, 0)) {
+    /* node0 is not DRAM, and we assume that's not possible */
+    hwloc_debug("    node0 shouldn't have HBM BW\n");
+    return -1;
+  }
+
+  /* assume tier1 == DRAM and tier2 == HBM */
+  tier1->type = HWLOC_MEMORY_TIER_DRAM;
+  tier2->type = HWLOC_MEMORY_TIER_HBM;
+  hwloc_debug("    Success\n");
+  return 0;
+}
+
+static int
+hwloc__guess_memory_tiers_types(hwloc_topology_t topology __hwloc_attribute_unused,
+                                unsigned nr_tiers,
+                                struct hwloc_memory_tier_s *tiers)
+{
+  unsigned long flags;
+  const char *env;
+  unsigned nr_unknown, nr_spm;
+  struct hwloc_memory_tier_s *unknown_tier[2], *spm_tier;
+  unsigned i;
+
+  flags = 0;
+  env = getenv("HWLOC_MEMTIERS_GUESS");
+  if (env) {
+    if (!strcmp(env, "none"))
+      return 0;
+    /* by default, we don't guess anything unsure */
+    if (!strcmp(env, "all"))
+      /* enable all typical cases */
+      flags = ~0UL;
+    if (strstr(env, "spm_is_hbm")) {
+      hwloc_debug("Assuming SPM-tier is HBM, ignore bandwidth\n");
+      flags |= HWLOC_GUESS_MEMTIERS_FLAG_SPM_IS_HBM;
+    }
+    if (strstr(env, "node0_is_dram")) {
+      hwloc_debug("Assuming node0 is DRAM\n");
+      flags |= HWLOC_GUESS_MEMTIERS_FLAG_NODE0_IS_DRAM;
+    }
+  }
+
+  if (nr_tiers == 1)
+    /* Likely DRAM only, but could also be HBM-only in non-SPM mode.
+     * We cannot be sure, but it doesn't matter since there's a single tier.
+     */
+    return 0;
+
+  nr_unknown = nr_spm = 0;
+  unknown_tier[0] = unknown_tier[1] = spm_tier = NULL;
+  for(i=0; i<nr_tiers; i++) {
+    switch (tiers[i].type) {
+    case HWLOC_MEMORY_TIER_UNKNOWN:
+      if (nr_unknown < 2)
+        unknown_tier[nr_unknown] = &tiers[i];
+      nr_unknown++;
+      break;
+    case HWLOC_MEMORY_TIER_SPM:
+      spm_tier = &tiers[i];
+      nr_spm++;
+      break;
+    case HWLOC_MEMORY_TIER_DRAM:
+    case HWLOC_MEMORY_TIER_HBM:
+      /* not possible */
+      abort();
+    default:
+      /* ignore HBM, NVM, ... */
+      break;
+    }
+  }
+  hwloc_debug("Found %u unknown memory tiers and %u SPM\n",
+              nr_unknown, nr_spm);
+
+  /* Try to guess DRAM + HBM common cases.
+   * Other things we'd like to detect:
+   * single unknown => DRAM or HBM? HBM won't be SPM on HBM-only CPUs
+   * unknown + CXL DRAM => DRAM or HBM?
+   */
+  if (nr_unknown == 2 && !nr_spm) {
+    /* 2 unknown, could be DRAM + non-SPM HBM */
+    hwloc_debug("  Trying to guess 2 unknown tiers using BW\n");
+    hwloc__guess_dram_hbm_tiers(unknown_tier[0], unknown_tier[1], flags);
+  } else if (nr_unknown == 1 && nr_spm == 1) {
+    /* 1 unknown + 1 SPM, could be DRAM + SPM HBM */
+    hwloc_debug("  Trying to guess 1 unknown + 1 SPM tiers using BW\n");
+    hwloc__guess_dram_hbm_tiers(unknown_tier[0], spm_tier, flags);
+  }
+
+  if (flags & HWLOC_GUESS_MEMTIERS_FLAG_SPM_IS_HBM) {
+    /* force mark SPM as HBM */
+    for(i=0; i<nr_tiers; i++)
+      if (tiers[i].type == HWLOC_MEMORY_TIER_SPM) {
+        hwloc_debug("Forcing SPM tier to HBM");
+        tiers[i].type = HWLOC_MEMORY_TIER_HBM;
+      }
+  }
+
+  if (flags & HWLOC_GUESS_MEMTIERS_FLAG_NODE0_IS_DRAM) {
+    /* force mark node0's tier as DRAM if we couldn't guess it */
+    for(i=0; i<nr_tiers; i++)
+      if (hwloc_bitmap_isset(tiers[i].nodeset, 0)
+          && tiers[i].type == HWLOC_MEMORY_TIER_UNKNOWN) {
+        hwloc_debug("Forcing node0 tier to DRAM");
+        tiers[i].type = HWLOC_MEMORY_TIER_DRAM;
+        break;
+      }
+  }
+
+  return 0;
+}
+
+/* parses something like 0xf=HBM;0x0f=DRAM;0x00f=CXL-DRAM */
+static struct hwloc_memory_tier_s *
+hwloc__force_memory_tiers(hwloc_topology_t topology __hwloc_attribute_unused,
+                          unsigned *nr_tiers_p,
+                          const char *_env)
+{
+  struct hwloc_memory_tier_s *tiers = NULL;
+  unsigned nr_tiers, i;
+  hwloc_bitmap_t nodeset = NULL;
+  char *env;
+  const char *tmp;
+
+  env = strdup(_env);
+  if (!env) {
+    fprintf(stderr, "[hwloc/memtiers] failed to duplicate HWLOC_MEMTIERS envvar\n");
+    goto out;
+  }
+
+  tmp = env;
+  nr_tiers = 1;
+  while (1) {
+    tmp = strchr(tmp, ';');
+    if (!tmp)
+      break;
+    tmp++;
+    nr_tiers++;
+  }
+
+  nodeset = hwloc_bitmap_alloc();
+  if (!nodeset) {
+    fprintf(stderr, "[hwloc/memtiers] failed to allocated forced tiers' nodeset\n");
+    goto out_with_envvar;
+  }
+
+  tiers = calloc(nr_tiers, sizeof(*tiers));
+  if (!tiers) {
+    fprintf(stderr, "[hwloc/memtiers] failed to allocated forced tiers\n");
+    goto out_with_nodeset;
+  }
+  nr_tiers = 0;
+
+  tmp = env;
+  while (1) {
+    char *end;
+    char *equal;
+    hwloc_memory_tier_type_t type;
+
+    end = strchr(tmp, ';');
+    if (end)
+      *end = '\0';
+
+    equal = strchr(tmp, '=');
+    if (!equal) {
+      fprintf(stderr, "[hwloc/memtiers] missing `=' before end of forced tier description at `%s'\n", tmp);
+      goto out_with_tiers;
+    }
+    *equal = '\0';
+
+    hwloc_bitmap_sscanf(nodeset, tmp);
+    if (hwloc_bitmap_iszero(nodeset)) {
+      fprintf(stderr, "[hwloc/memtiers] empty forced tier nodeset `%s', aborting\n", tmp);
+      goto out_with_tiers;
+    }
+    type = hwloc_memory_tier_type_sscanf(equal+1);
+    if (!type)
+      hwloc_debug("failed to recognize forced tier type `%s'\n", equal+1);
+    tiers[nr_tiers].nodeset = hwloc_bitmap_dup(nodeset);
+    tiers[nr_tiers].type = type;
+    tiers[nr_tiers].local_bw_min = tiers[nr_tiers].local_bw_max = 0;
+    tiers[nr_tiers].local_lat_min = tiers[nr_tiers].local_lat_max = 0;
+    nr_tiers++;
+    if (!end)
+      break;
+    tmp = end+1;
+  }
+
+  free(env);
+  hwloc_bitmap_free(nodeset);
+  hwloc_debug("Forcing %u memory tiers\n", nr_tiers);
+#ifdef HWLOC_DEBUG
+  for(i=0; i<nr_tiers; i++) {
+    char *s;
+    hwloc_bitmap_asprintf(&s, tiers[i].nodeset);
+    hwloc_debug("  tier #%u type %lx nodeset %s\n", i, tiers[i].type, s);
+    free(s);
+  }
+#endif
+  *nr_tiers_p = nr_tiers;
+  return tiers;
+
+ out_with_tiers:
+  for(i=0; i<nr_tiers; i++)
+    hwloc_bitmap_free(tiers[i].nodeset);
+  free(tiers);
+ out_with_nodeset:
+  hwloc_bitmap_free(nodeset);
+ out_with_envvar:
+  free(env);
+ out:
+  return NULL;
+}
+
+static void
+hwloc__apply_memory_tiers_subtypes(hwloc_topology_t topology,
+                                   unsigned nr_tiers,
+                                   struct hwloc_memory_tier_s *tiers,
+                                   int force)
+{
+  hwloc_obj_t node = NULL;
+  hwloc_debug("Marking node tiers\n");
+  while ((node = hwloc_get_next_obj_by_type(topology, HWLOC_OBJ_NUMANODE, node)) != NULL) {
+    unsigned j;
+    for(j=0; j<nr_tiers; j++) {
+      if (hwloc_bitmap_isset(tiers[j].nodeset, node->os_index)) {
+        const char *subtype = hwloc_memory_tier_type_snprintf(tiers[j].type);
+        if (!node->subtype || force) { /* don't overwrite the existing subtype unless forced */
+          if (subtype) { /* don't set a subtype for unknown tiers */
+            hwloc_debug("  marking node L#%u P#%u as %s (was %s)\n", node->logical_index, node->os_index, subtype, node->subtype);
+            free(node->subtype);
+            node->subtype = strdup(subtype);
+          }
+        } else
+          hwloc_debug("  node L#%u P#%u already marked as %s, not setting %s\n",
+                      node->logical_index, node->os_index, node->subtype, subtype);
+        if (nr_tiers > 1) {
+          char tmp[20];
+          snprintf(tmp, sizeof(tmp), "%u", j);
+          hwloc__add_info_nodup(&node->infos, &node->infos_count, "MemoryTier", tmp, 1);
+        }
+        break; /* each node is in a single tier */
+      }
+    }
+  }
+  if (nr_tiers > 1) {
+    hwloc_obj_t root = hwloc_get_root_obj(topology);
+    char tmp[20];
+    snprintf(tmp, sizeof(tmp), "%u", nr_tiers);
+    hwloc__add_info_nodup(&root->infos, &root->infos_count, "MemoryTiersNr", tmp, 1);
+  }
+}
+
+int
+hwloc_internal_memattrs_guess_memory_tiers(hwloc_topology_t topology, int force_subtype)
+{
+  struct hwloc_memory_tier_s *tiers;
+  unsigned nr_tiers;
+  unsigned i;
+  const char *env;
+
+  env = getenv("HWLOC_MEMTIERS");
+  if (env) {
+    if (!strcmp(env, "none"))
+      goto out;
+    tiers = hwloc__force_memory_tiers(topology, &nr_tiers, env);
+    if (tiers) {
+      assert(nr_tiers > 0);
+      force_subtype = 1;
+      goto ready;
+    }
+  }
+
+  tiers = hwloc__group_memory_tiers(topology, &nr_tiers);
+  if (!tiers)
+    goto out;
+
+  hwloc__guess_memory_tiers_types(topology, nr_tiers, tiers);
+
+  /* sort tiers by BW first, then by type */
+  hwloc_debug("Sorting memory tiers...\n");
+  qsort(tiers, nr_tiers, sizeof(*tiers), compare_tiers_by_bw_and_type);
+
+ ready:
+#ifdef HWLOC_DEBUG
+  for(i=0; i<nr_tiers; i++) {
+    char *s;
+    hwloc_bitmap_asprintf(&s, tiers[i].nodeset);
+    hwloc_debug("  tier %u = nodes %s with type %lx and local BW %llu-%llu lat %llu-%llu\n",
+                i,
+                s, tiers[i].type,
+                (unsigned long long) tiers[i].local_bw_min,
+                (unsigned long long) tiers[i].local_bw_max,
+                (unsigned long long) tiers[i].local_lat_min,
+                (unsigned long long) tiers[i].local_lat_max);
+    free(s);
+  }
+#endif
+
+  hwloc__apply_memory_tiers_subtypes(topology, nr_tiers, tiers, force_subtype);
+
+  for(i=0; i<nr_tiers; i++)
+    hwloc_bitmap_free(tiers[i].nodeset);
+  free(tiers);
+ out:
+  return 0;
+}

src/3rdparty/hwloc/src/pci-common.c

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2024 Inria.  All rights reserved.
  * See COPYING in top-level directory.
  */
 
@@ -119,6 +119,13 @@ hwloc_pci_discovery_init(struct hwloc_topology *topology)
   topology->pci_forced_locality = NULL;
 
   topology->first_pci_locality = topology->last_pci_locality = NULL;
+
+#define HWLOC_PCI_LOCALITY_QUIRK_CRAY_EX235A (1ULL<<0)
+#define HWLOC_PCI_LOCALITY_QUIRK_FAKE (1ULL<<62)
+  topology->pci_locality_quirks = (uint64_t) -1;
+  /* -1 is unknown, 0 is disabled, >0 is bitmask of enabled quirks.
+   * bit 63 should remain unused so that -1 is unaccessible as a bitmask.
+   */
 }
 
 void
@@ -146,8 +153,9 @@ hwloc_pci_discovery_prepare(struct hwloc_topology *topology)
 	  }
 	  free(buffer);
 	} else {
-	  fprintf(stderr, "Ignoring HWLOC_PCI_LOCALITY file `%s' too large (%lu bytes)\n",
+          if (HWLOC_SHOW_CRITICAL_ERRORS())
-		  env, (unsigned long) st.st_size);
+            fprintf(stderr, "hwloc/pci: Ignoring HWLOC_PCI_LOCALITY file `%s' too large (%lu bytes)\n",
+                    env, (unsigned long) st.st_size);
 	}
       }
       close(fd);
@@ -206,8 +214,11 @@ hwloc_pci_traverse_print_cb(void * cbdata __hwloc_attribute_unused,
     else
       hwloc_debug("%s Bridge [%04x:%04x]", busid,
 		  pcidev->attr->pcidev.vendor_id, pcidev->attr->pcidev.device_id);
-    hwloc_debug(" to %04x:[%02x:%02x]\n",
+    if (pcidev->attr->bridge.downstream_type == HWLOC_OBJ_BRIDGE_PCI)
-		pcidev->attr->bridge.downstream.pci.domain, pcidev->attr->bridge.downstream.pci.secondary_bus, pcidev->attr->bridge.downstream.pci.subordinate_bus);
+      hwloc_debug(" to %04x:[%02x:%02x]\n",
+                  pcidev->attr->bridge.downstream.pci.domain, pcidev->attr->bridge.downstream.pci.secondary_bus, pcidev->attr->bridge.downstream.pci.subordinate_bus);
+    else
+      assert(0);
   } else
     hwloc_debug("%s Device [%04x:%04x (%04x:%04x) rev=%02x class=%04x]\n", busid,
 		pcidev->attr->pcidev.vendor_id, pcidev->attr->pcidev.device_id,
@@ -251,11 +262,11 @@ hwloc_pci_compare_busids(struct hwloc_obj *a, struct hwloc_obj *b)
   if (a->attr->pcidev.domain > b->attr->pcidev.domain)
     return HWLOC_PCI_BUSID_HIGHER;
 
-  if (a->type == HWLOC_OBJ_BRIDGE
+  if (a->type == HWLOC_OBJ_BRIDGE && a->attr->bridge.downstream_type == HWLOC_OBJ_BRIDGE_PCI
       && b->attr->pcidev.bus >= a->attr->bridge.downstream.pci.secondary_bus
       && b->attr->pcidev.bus <= a->attr->bridge.downstream.pci.subordinate_bus)
     return HWLOC_PCI_BUSID_SUPERSET;
-  if (b->type == HWLOC_OBJ_BRIDGE
+  if (b->type == HWLOC_OBJ_BRIDGE && b->attr->bridge.downstream_type == HWLOC_OBJ_BRIDGE_PCI
       && a->attr->pcidev.bus >= b->attr->bridge.downstream.pci.secondary_bus
       && a->attr->pcidev.bus <= b->attr->bridge.downstream.pci.subordinate_bus)
     return HWLOC_PCI_BUSID_INCLUDED;
@@ -302,7 +313,7 @@ hwloc_pci_add_object(struct hwloc_obj *parent, struct hwloc_obj **parent_io_firs
       new->next_sibling = *curp;
       *curp = new;
       new->parent = parent;
-      if (new->type == HWLOC_OBJ_BRIDGE) {
+      if (new->type == HWLOC_OBJ_BRIDGE && new->attr->bridge.downstream_type == HWLOC_OBJ_BRIDGE_PCI) {
 	/* look at remaining siblings and move some below new */
 	childp = &new->io_first_child;
 	curp = &new->next_sibling;
@@ -329,7 +340,7 @@ hwloc_pci_add_object(struct hwloc_obj *parent, struct hwloc_obj **parent_io_firs
     }
     case HWLOC_PCI_BUSID_EQUAL: {
       static int reported = 0;
-      if (!reported && !hwloc_hide_errors()) {
+      if (!reported && HWLOC_SHOW_CRITICAL_ERRORS()) {
         fprintf(stderr, "*********************************************************\n");
         fprintf(stderr, "* hwloc %s received invalid PCI information.\n", HWLOC_VERSION);
         fprintf(stderr, "*\n");
@@ -411,7 +422,7 @@ hwloc_pcidisc_add_hostbridges(struct hwloc_topology *topology,
     dstnextp = &child->next_sibling;
 
     /* compute hostbridge secondary/subordinate buses */
-    if (child->type == HWLOC_OBJ_BRIDGE
+    if (child->type == HWLOC_OBJ_BRIDGE && child->attr->bridge.downstream_type == HWLOC_OBJ_BRIDGE_PCI
 	&& child->attr->bridge.downstream.pci.subordinate_bus > current_subordinate)
       current_subordinate = child->attr->bridge.downstream.pci.subordinate_bus;
 
@@ -438,13 +449,90 @@ hwloc_pcidisc_add_hostbridges(struct hwloc_topology *topology,
   return new;
 }
 
-static struct hwloc_obj *
+/* return 1 if a quirk was applied */
-hwloc_pci_fixup_busid_parent(struct hwloc_topology *topology __hwloc_attribute_unused,
+static int
-			     struct hwloc_pcidev_attr_s *busid __hwloc_attribute_unused,
+hwloc__pci_find_busid_parent_quirk(struct hwloc_topology *topology,
-			     struct hwloc_obj *parent __hwloc_attribute_unused)
+                                   struct hwloc_pcidev_attr_s *busid,
+                                   hwloc_cpuset_t cpuset)
 {
-  /* no quirk for now */
+  if (topology->pci_locality_quirks == (uint64_t)-1 /* unknown */) {
-  return parent;
+    const char *dmi_board_name, *env;
+
+    /* first invokation, detect which quirks are needed */
+    topology->pci_locality_quirks = 0; /* no quirk yet */
+
+    dmi_board_name = hwloc_obj_get_info_by_name(hwloc_get_root_obj(topology), "DMIBoardName");
+    if (dmi_board_name && !strcmp(dmi_board_name, "HPE CRAY EX235A")) {
+      hwloc_debug("enabling for PCI locality quirk for HPE Cray EX235A\n");
+      topology->pci_locality_quirks |= HWLOC_PCI_LOCALITY_QUIRK_CRAY_EX235A;
+    }
+
+    env = getenv("HWLOC_PCI_LOCALITY_QUIRK_FAKE");
+    if (env && atoi(env)) {
+      hwloc_debug("enabling for PCI locality fake quirk (attaching everything to last PU)\n");
+      topology->pci_locality_quirks |= HWLOC_PCI_LOCALITY_QUIRK_FAKE;
+    }
+  }
+
+  if (topology->pci_locality_quirks & HWLOC_PCI_LOCALITY_QUIRK_FAKE) {
+    unsigned last = hwloc_bitmap_last(hwloc_topology_get_topology_cpuset(topology));
+    hwloc_bitmap_set(cpuset, last);
+    return 1;
+  }
+
+  if (topology->pci_locality_quirks & HWLOC_PCI_LOCALITY_QUIRK_CRAY_EX235A) {
+    /* AMD Trento has xGMI ports connected to individual CCDs (8 cores + L3)
+     * instead of NUMA nodes (pairs of CCDs within Trento) as is usual in AMD EPYC CPUs.
+     * This is not described by the ACPI tables, hence we need to manually hardwire
+     * the xGMI locality for the (currently single) server that currently uses that CPU.
+     * It's not clear if ACPI tables can/will ever be fixed (would require one initiator
+     * proximity domain per CCD), or if Linux can/will work around the issue.
+     */
+    if (busid->domain == 0) {
+      if (busid->bus >= 0xd0 && busid->bus <= 0xd1) {
+        hwloc_bitmap_set_range(cpuset, 0, 7);
+        hwloc_bitmap_set_range(cpuset, 64, 71);
+        return 1;
+      }
+      if (busid->bus >= 0xd4 && busid->bus <= 0xd6) {
+        hwloc_bitmap_set_range(cpuset, 8, 15);
+        hwloc_bitmap_set_range(cpuset, 72, 79);
+        return 1;
+      }
+      if (busid->bus >= 0xc8 && busid->bus <= 0xc9) {
+        hwloc_bitmap_set_range(cpuset, 16, 23);
+        hwloc_bitmap_set_range(cpuset, 80, 87);
+        return 1;
+      }
+      if (busid->bus >= 0xcc && busid->bus <= 0xce) {
+        hwloc_bitmap_set_range(cpuset, 24, 31);
+        hwloc_bitmap_set_range(cpuset, 88, 95);
+        return 1;
+      }
+      if (busid->bus >= 0xd8 && busid->bus <= 0xd9) {
+        hwloc_bitmap_set_range(cpuset, 32, 39);
+        hwloc_bitmap_set_range(cpuset, 96, 103);
+        return 1;
+      }
+      if (busid->bus >= 0xdc && busid->bus <= 0xde) {
+        hwloc_bitmap_set_range(cpuset, 40, 47);
+        hwloc_bitmap_set_range(cpuset, 104, 111);
+        return 1;
+      }
+      if (busid->bus >= 0xc0 && busid->bus <= 0xc1) {
+        hwloc_bitmap_set_range(cpuset, 48, 55);
+        hwloc_bitmap_set_range(cpuset, 112, 119);
+        return 1;
+      }
+      if (busid->bus >= 0xc4 && busid->bus <= 0xc6) {
+        hwloc_bitmap_set_range(cpuset, 56, 63);
+        hwloc_bitmap_set_range(cpuset, 120, 127);
+        return 1;
+      }
+    }
+  }
+
+  return 0;
 }
 
 static struct hwloc_obj *
@@ -453,7 +541,7 @@ hwloc__pci_find_busid_parent(struct hwloc_topology *topology, struct hwloc_pcide
   hwloc_bitmap_t cpuset = hwloc_bitmap_alloc();
   hwloc_obj_t parent;
   int forced = 0;
-  int noquirks = 0;
+  int noquirks = 0, got_quirked = 0;
   unsigned i;
   int err;
 
@@ -486,7 +574,8 @@ hwloc__pci_find_busid_parent(struct hwloc_topology *topology, struct hwloc_pcide
     if (env) {
       static int reported = 0;
       if (!topology->pci_has_forced_locality && !reported) {
-	fprintf(stderr, "Environment variable %s is deprecated, please use HWLOC_PCI_LOCALITY instead.\n", env);
+        if (HWLOC_SHOW_ALL_ERRORS())
+          fprintf(stderr, "hwloc/pci: Environment variable %s is deprecated, please use HWLOC_PCI_LOCALITY instead.\n", env);
 	reported = 1;
       }
       if (*env) {
@@ -500,7 +589,13 @@ hwloc__pci_find_busid_parent(struct hwloc_topology *topology, struct hwloc_pcide
     }
   }
 
-  if (!forced) {
+  if (!forced && !noquirks && topology->pci_locality_quirks /* either quirks are unknown yet, or some are enabled */) {
+    err = hwloc__pci_find_busid_parent_quirk(topology, busid, cpuset);
+    if (err > 0)
+      got_quirked = 1;
+  }
+
+  if (!forced && !got_quirked) {
     /* get the cpuset by asking the backend that provides the relevant hook, if any. */
     struct hwloc_backend *backend = topology->get_pci_busid_cpuset_backend;
     if (backend)
@@ -515,11 +610,7 @@ hwloc__pci_find_busid_parent(struct hwloc_topology *topology, struct hwloc_pcide
   hwloc_debug_bitmap("  will attach PCI bus to cpuset %s\n", cpuset);
 
   parent = hwloc_find_insert_io_parent_by_complete_cpuset(topology, cpuset);
-  if (parent) {
+  if (!parent) {
-    if (!noquirks)
-      /* We found a valid parent. Check that the OS didn't report invalid locality */
-      parent = hwloc_pci_fixup_busid_parent(topology, busid, parent);
-  } else {
     /* Fallback to root */
     parent = hwloc_get_root_obj(topology);
   }
@@ -565,7 +656,7 @@ hwloc_pcidisc_tree_attach(struct hwloc_topology *topology, struct hwloc_obj *tre
     assert(pciobj->type == HWLOC_OBJ_PCI_DEVICE
 	   || (pciobj->type == HWLOC_OBJ_BRIDGE && pciobj->attr->bridge.upstream_type == HWLOC_OBJ_BRIDGE_PCI));
 
-    if (obj->type == HWLOC_OBJ_BRIDGE) {
+    if (obj->type == HWLOC_OBJ_BRIDGE && obj->attr->bridge.downstream_type == HWLOC_OBJ_BRIDGE_PCI) {
       domain = obj->attr->bridge.downstream.pci.domain;
       bus_min = obj->attr->bridge.downstream.pci.secondary_bus;
       bus_max = obj->attr->bridge.downstream.pci.subordinate_bus;
@@ -795,26 +886,12 @@ hwloc_pcidisc_find_linkspeed(const unsigned char *config,
 			     unsigned offset, float *linkspeed)
 {
   unsigned linksta, speed, width;
-  float lanespeed;
 
   memcpy(&linksta, &config[offset + HWLOC_PCI_EXP_LNKSTA], 4);
   speed = linksta & HWLOC_PCI_EXP_LNKSTA_SPEED; /* PCIe generation */
   width = (linksta & HWLOC_PCI_EXP_LNKSTA_WIDTH) >> 4; /* how many lanes */
-  /* PCIe Gen1 = 2.5GT/s signal-rate per lane with 8/10 encoding    = 0.25GB/s data-rate per lane
-   * PCIe Gen2 = 5  GT/s signal-rate per lane with 8/10 encoding    = 0.5 GB/s data-rate per lane
-   * PCIe Gen3 = 8  GT/s signal-rate per lane with 128/130 encoding = 1   GB/s data-rate per lane
-   * PCIe Gen4 = 16 GT/s signal-rate per lane with 128/130 encoding = 2   GB/s data-rate per lane
-   * PCIe Gen5 = 32 GT/s signal-rate per lane with 128/130 encoding = 4   GB/s data-rate per lane
-   */
 
-  /* lanespeed in Gbit/s */
+  *linkspeed = hwloc__pci_link_speed(speed, width);
-  if (speed <= 2)
-    lanespeed = 2.5f * speed * 0.8f;
-  else
-    lanespeed = 8.0f * (1<<(speed-3)) * 128/130; /* assume Gen6 will be 64 GT/s and so on */
-
-  /* linkspeed in GB/s */
-  *linkspeed = lanespeed * width / 8;
   return 0;
 }
 
@@ -938,6 +1015,7 @@ hwloc_pci_class_string(unsigned short class_id)
       switch (class_id) {
 	case 0x0500: return "RAM";
 	case 0x0501: return "Flash";
+        case 0x0502: return "CXLMem";
       }
       return "Memory";
     case 0x06:

src/3rdparty/hwloc/src/shmem.c

@@ -23,6 +23,7 @@ struct hwloc_shmem_header {
   uint32_t header_length; /* where the actual topology starts in the file/mapping */
   uint64_t mmap_address; /* virtual address to pass to mmap */
   uint64_t mmap_length; /* length to pass to mmap (includes the header) */
+  /* we will pad the end to a multiple of pointer size so that the topology is well aligned */
 };
 
 #define HWLOC_SHMEM_MALLOC_ALIGN 8UL
@@ -85,6 +86,7 @@ hwloc_shmem_topology_write(hwloc_topology_t topology,
   hwloc_topology_t new;
   struct hwloc_tma tma;
   struct hwloc_shmem_header header;
+  uint32_t header_length = (sizeof(header) + sizeof(void*) - 1) & ~(sizeof(void*) - 1); /* pad to a multiple of pointer size */
   void *mmap_res;
   int err;
 
@@ -100,7 +102,7 @@ hwloc_shmem_topology_write(hwloc_topology_t topology,
   hwloc_internal_memattrs_refresh(topology);
 
   header.header_version = HWLOC_SHMEM_HEADER_VERSION;
-  header.header_length = sizeof(header);
+  header.header_length = header_length;
   header.mmap_address = (uintptr_t) mmap_address;
   header.mmap_length = length;
 
@@ -127,7 +129,7 @@ hwloc_shmem_topology_write(hwloc_topology_t topology,
 
   tma.malloc = tma_shmem_malloc;
   tma.dontfree = 1;
-  tma.data = (char *)mmap_res + sizeof(header);
+  tma.data = (char *)mmap_res + header_length;
   err = hwloc__topology_dup(&new, topology, &tma);
   if (err < 0)
     return err;
@@ -154,6 +156,7 @@ hwloc_shmem_topology_adopt(hwloc_topology_t *topologyp,
 {
   hwloc_topology_t new, old;
   struct hwloc_shmem_header header;
+  uint32_t header_length = (sizeof(header) + sizeof(void*) - 1) & ~(sizeof(void*) - 1); /* pad to a multiple of pointer size */
   void *mmap_res;
   int err;
 
@@ -171,7 +174,7 @@ hwloc_shmem_topology_adopt(hwloc_topology_t *topologyp,
     return -1;
 
   if (header.header_version != HWLOC_SHMEM_HEADER_VERSION
-      || header.header_length != sizeof(header)
+      || header.header_length != header_length
       || header.mmap_address != (uintptr_t) mmap_address
       || header.mmap_length != length) {
     errno = EINVAL;
@@ -186,7 +189,7 @@ hwloc_shmem_topology_adopt(hwloc_topology_t *topologyp,
     goto out_with_mmap;
   }
 
-  old = (hwloc_topology_t)((char*)mmap_address + sizeof(header));
+  old = (hwloc_topology_t)((char*)mmap_address + header_length);
   if (hwloc_topology_abi_check(old) < 0) {
     errno = EINVAL;
     goto out_with_mmap;

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

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

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2023 Inria.  All rights reserved.
  * Copyright © 2009-2010 Université Bordeaux
  * Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -23,6 +23,7 @@ struct hwloc_synthetic_attr_s {
   unsigned depth; /* For caches/groups */
   hwloc_obj_cache_type_t cachetype; /* For caches */
   hwloc_uint64_t memorysize; /* For caches/memory */
+  hwloc_uint64_t memorysidecachesize; /* Single level of memory-side-cache in-front of a NUMA node */
 };
 
 struct hwloc_synthetic_indexes_s {
@@ -323,17 +324,29 @@ hwloc_synthetic_parse_memory_attr(const char *attr, const char **endp)
   hwloc_uint64_t size;
   size = strtoull(attr, (char **) &endptr, 0);
   if (!hwloc_strncasecmp(endptr, "TB", 2)) {
+    size *= 1000ULL*1000ULL*1000ULL*1000ULL;
+    endptr += 2;
+  } else if (!hwloc_strncasecmp(endptr, "TiB", 3)) {
     size <<= 40;
-    endptr += 2;
+    endptr += 3;
   } else if (!hwloc_strncasecmp(endptr, "GB", 2)) {
+    size *= 1000ULL*1000ULL*1000ULL;
+    endptr += 2;
+  } else if (!hwloc_strncasecmp(endptr, "GiB", 3)) {
     size <<= 30;
-    endptr += 2;
+    endptr += 3;
   } else if (!hwloc_strncasecmp(endptr, "MB", 2)) {
+    size *= 1000ULL*1000ULL;
+    endptr += 2;
+  } else if (!hwloc_strncasecmp(endptr, "MiB", 3)) {
     size <<= 20;
-    endptr += 2;
+    endptr += 3;
   } else if (!hwloc_strncasecmp(endptr, "kB", 2)) {
-    size <<= 10;
+    size *= 1000ULL;
     endptr += 2;
+  } else if (!hwloc_strncasecmp(endptr, "kiB", 3)) {
+    size <<= 10;
+    endptr += 3;
   }
   *endp = endptr;
   return size;
@@ -368,6 +381,9 @@ hwloc_synthetic_parse_attrs(const char *attrs, const char **next_posp,
     } else if (!iscache && !strncmp("memory=", attrs, 7)) {
       memorysize = hwloc_synthetic_parse_memory_attr(attrs+7, &attrs);
 
+    } else if (!strncmp("memorysidecachesize=", attrs, 20)) {
+      sattr->memorysidecachesize = hwloc_synthetic_parse_memory_attr(attrs+20, &attrs);
+
     } else if (!strncmp("indexes=", attrs, 8)) {
       index_string = attrs+8;
       attrs += 8;
@@ -375,10 +391,9 @@ hwloc_synthetic_parse_attrs(const char *attrs, const char **next_posp,
       attrs += index_string_length;
 
     } else {
-      if (verbose)
+      size_t length = strcspn(attrs, " )");
-	fprintf(stderr, "Unknown attribute at '%s'\n", attrs);
+      fprintf(stderr, "hwloc/synthetic: Ignoring unknown attribute at '%s'\n", attrs);
-      errno = EINVAL;
+      attrs += length;
-      return -1;
     }
 
     if (' ' == *attrs)
@@ -404,6 +419,32 @@ hwloc_synthetic_parse_attrs(const char *attrs, const char **next_posp,
   return 0;
 }
 
+static void
+hwloc_synthetic_set_default_attrs(struct hwloc_synthetic_attr_s *sattr,
+                                  int *type_count)
+{
+  hwloc_obj_type_t type = sattr->type;
+
+  if (type == HWLOC_OBJ_GROUP) {
+    if (sattr->depth == (unsigned)-1)
+      sattr->depth = type_count[HWLOC_OBJ_GROUP]--;
+
+  } else if (hwloc__obj_type_is_cache(type)) {
+    if (!sattr->memorysize) {
+      if (1 == sattr->depth)
+        /* 32KiB in L1 */
+        sattr->memorysize = 32*1024;
+      else
+        /* *4 at each level, starting from 1MiB for L2, unified */
+        sattr->memorysize = 256ULL*1024 << (2*sattr->depth);
+    }
+
+  } else if (type == HWLOC_OBJ_NUMANODE && !sattr->memorysize) {
+    /* 1GiB in memory nodes. */
+    sattr->memorysize = 1024*1024*1024;
+  }
+}
+
 /* frees level until arity = 0 */
 static void
 hwloc_synthetic_free_levels(struct hwloc_synthetic_backend_data_s *data)
@@ -453,6 +494,7 @@ hwloc_backend_synthetic_init(struct hwloc_synthetic_backend_data_s *data,
   data->level[0].indexes.string = NULL;
   data->level[0].indexes.array = NULL;
   data->level[0].attr.memorysize = 0;
+  data->level[0].attr.memorysidecachesize = 0;
   data->level[0].attached = NULL;
   type_count[HWLOC_OBJ_MACHINE] = 1;
   if (*description == '(') {
@@ -502,6 +544,7 @@ hwloc_backend_synthetic_init(struct hwloc_synthetic_backend_data_s *data,
       if (attached) {
 	attached->attr.type = type;
 	attached->attr.memorysize = 0;
+	attached->attr.memorysidecachesize = 0;
 	/* attached->attr.depth and .cachetype unused */
 	attached->next = NULL;
 	pprev = &data->level[count-1].attached;
@@ -589,7 +632,7 @@ hwloc_backend_synthetic_init(struct hwloc_synthetic_backend_data_s *data,
     }
     if (!item) {
       if (verbose)
-	fprintf(stderr,"Synthetic string with disallow 0 number of objects at '%s'\n", pos);
+	fprintf(stderr,"Synthetic string with disallowed 0 number of objects at '%s'\n", pos);
       errno = EINVAL;
       goto error;
     }
@@ -599,6 +642,7 @@ hwloc_backend_synthetic_init(struct hwloc_synthetic_backend_data_s *data,
     data->level[count].indexes.string = NULL;
     data->level[count].indexes.array = NULL;
     data->level[count].attr.memorysize = 0;
+    data->level[count].attr.memorysidecachesize = 0;
     if (*next_pos == '(') {
       err = hwloc_synthetic_parse_attrs(next_pos+1, &next_pos, &data->level[count].attr, &data->level[count].indexes, verbose);
       if (err < 0)
@@ -784,6 +828,7 @@ hwloc_backend_synthetic_init(struct hwloc_synthetic_backend_data_s *data,
     data->level[1].indexes.string = NULL;
     data->level[1].indexes.array = NULL;
     data->level[1].attr.memorysize = 0;
+    data->level[1].attr.memorysidecachesize = 0;
     data->level[1].totalwidth = data->level[0].totalwidth;
     /* update arity to insert a single NUMA node per parent */
     data->level[1].arity = data->level[0].arity;
@@ -791,30 +836,14 @@ hwloc_backend_synthetic_init(struct hwloc_synthetic_backend_data_s *data,
     count++;
   }
 
+  /* set default attributes that depend on the depth/hierarchy of levels */
   for (i=0; i<count; i++) {
+    struct hwloc_synthetic_attached_s *attached;
     struct hwloc_synthetic_level_data_s *curlevel = &data->level[i];
-    hwloc_obj_type_t type = curlevel->attr.type;
+    hwloc_synthetic_set_default_attrs(&curlevel->attr, type_count);
-
+    for(attached = curlevel->attached; attached != NULL; attached = attached->next)
-    if (type == HWLOC_OBJ_GROUP) {
+      hwloc_synthetic_set_default_attrs(&attached->attr, type_count);
-      if (curlevel->attr.depth == (unsigned)-1)
+    hwloc_synthetic_process_indexes(data, &curlevel->indexes, curlevel->totalwidth, verbose);
-	curlevel->attr.depth = type_count[HWLOC_OBJ_GROUP]--;
-
-    } else if (hwloc__obj_type_is_cache(type)) {
-      if (!curlevel->attr.memorysize) {
-	if (1 == curlevel->attr.depth)
-	  /* 32Kb in L1 */
-	  curlevel->attr.memorysize = 32*1024;
-	else
-	  /* *4 at each level, starting from 1MB for L2, unified */
-	  curlevel->attr.memorysize = 256ULL*1024 << (2*curlevel->attr.depth);
-      }
-
-    } else if (type == HWLOC_OBJ_NUMANODE && !curlevel->attr.memorysize) {
-      /* 1GB in memory nodes. */
-      curlevel->attr.memorysize = 1024*1024*1024;
-    }
-
-    hwloc_synthetic_process_indexes(data, &data->level[i].indexes, data->level[i].totalwidth, verbose);
   }
 
   hwloc_synthetic_process_indexes(data, &data->numa_attached_indexes, data->numa_attached_nr, verbose);
@@ -847,6 +876,12 @@ hwloc_synthetic_set_attr(struct hwloc_synthetic_attr_s *sattr,
     obj->attr->numanode.page_types[0].size = 4096;
     obj->attr->numanode.page_types[0].count = sattr->memorysize / 4096;
     break;
+  case HWLOC_OBJ_MEMCACHE:
+    obj->attr->cache.depth = 1;
+    obj->attr->cache.linesize = 64;
+    obj->attr->cache.type = HWLOC_OBJ_CACHE_UNIFIED;
+    obj->attr->cache.size = sattr->memorysidecachesize;
+    break;
   case HWLOC_OBJ_PACKAGE:
   case HWLOC_OBJ_DIE:
     break;
@@ -914,6 +949,14 @@ hwloc_synthetic_insert_attached(struct hwloc_topology *topology,
 
   hwloc__insert_object_by_cpuset(topology, NULL, child, "synthetic:attached");
 
+  if (attached->attr.memorysidecachesize) {
+    hwloc_obj_t mscachechild = hwloc_alloc_setup_object(topology, HWLOC_OBJ_MEMCACHE, HWLOC_UNKNOWN_INDEX);
+    mscachechild->cpuset = hwloc_bitmap_dup(set);
+    mscachechild->nodeset = hwloc_bitmap_dup(child->nodeset);
+    hwloc_synthetic_set_attr(&attached->attr, mscachechild);
+    hwloc__insert_object_by_cpuset(topology, NULL, mscachechild, "synthetic:attached:mscache");
+  }
+
   hwloc_synthetic_insert_attached(topology, data, attached->next, set);
 }
 
@@ -965,6 +1008,14 @@ hwloc__look_synthetic(struct hwloc_topology *topology,
     hwloc_synthetic_set_attr(&curlevel->attr, obj);
 
     hwloc__insert_object_by_cpuset(topology, NULL, obj, "synthetic");
+
+    if (type == HWLOC_OBJ_NUMANODE && curlevel->attr.memorysidecachesize) {
+      hwloc_obj_t mscachechild = hwloc_alloc_setup_object(topology, HWLOC_OBJ_MEMCACHE, HWLOC_UNKNOWN_INDEX);
+      mscachechild->cpuset = hwloc_bitmap_dup(set);
+      mscachechild->nodeset = hwloc_bitmap_dup(obj->nodeset);
+      hwloc_synthetic_set_attr(&curlevel->attr, mscachechild);
+      hwloc__insert_object_by_cpuset(topology, NULL, mscachechild, "synthetic:mscache");
+    }
   }
 
   hwloc_synthetic_insert_attached(topology, data, curlevel->attached, set);
@@ -1205,6 +1256,7 @@ hwloc__export_synthetic_indexes(hwloc_obj_t *level, unsigned total,
 
 static int
 hwloc__export_synthetic_obj_attr(struct hwloc_topology * topology,
+                                 unsigned long flags,
 				 hwloc_obj_t obj,
 				 char *buffer, size_t buflen)
 {
@@ -1212,6 +1264,7 @@ hwloc__export_synthetic_obj_attr(struct hwloc_topology * topology,
   const char * prefix = "(";
   char cachesize[64] = "";
   char memsize[64] = "";
+  char memorysidecachesize[64] = "";
   int needindexes = 0;
 
   if (hwloc__obj_type_is_cache(obj->type) && obj->attr->cache.size) {
@@ -1224,6 +1277,19 @@ hwloc__export_synthetic_obj_attr(struct hwloc_topology * topology,
 	     prefix, (unsigned long long) obj->attr->numanode.local_memory);
     prefix = separator;
   }
+  if (obj->type == HWLOC_OBJ_NUMANODE && !(flags & HWLOC_TOPOLOGY_EXPORT_SYNTHETIC_FLAG_V1)) {
+    hwloc_obj_t memorysidecache = obj->parent;
+    hwloc_uint64_t size = 0;
+    while (memorysidecache && memorysidecache->type == HWLOC_OBJ_MEMCACHE) {
+      size += memorysidecache->attr->cache.size;
+      memorysidecache = memorysidecache->parent;
+    }
+    if (size) {
+      snprintf(memorysidecachesize, sizeof(memorysidecachesize), "%smemorysidecachesize=%llu",
+               prefix, (unsigned long long) size);
+      prefix = separator;
+    }
+  }
   if (!obj->logical_index /* only display indexes once per level (not for non-first NUMA children, etc.) */
       && (obj->type == HWLOC_OBJ_PU || obj->type == HWLOC_OBJ_NUMANODE)) {
     hwloc_obj_t cur = obj;
@@ -1235,12 +1301,12 @@ hwloc__export_synthetic_obj_attr(struct hwloc_topology * topology,
       cur = cur->next_cousin;
     }
   }
-  if (*cachesize || *memsize || needindexes) {
+  if (*cachesize || *memsize || *memorysidecachesize || needindexes) {
     ssize_t tmplen = buflen;
     char *tmp = buffer;
     int res, ret = 0;
 
-    res = hwloc_snprintf(tmp, tmplen, "%s%s%s", cachesize, memsize, needindexes ? "" : ")");
+    res = hwloc_snprintf(tmp, tmplen, "%s%s%s%s", cachesize, memsize, memorysidecachesize, needindexes ? "" : ")");
     if (hwloc__export_synthetic_update_status(&ret, &tmp, &tmplen, res) < 0)
       return -1;
 
@@ -1314,7 +1380,7 @@ hwloc__export_synthetic_obj(struct hwloc_topology * topology, unsigned long flag
 
   if (!(flags & HWLOC_TOPOLOGY_EXPORT_SYNTHETIC_FLAG_NO_ATTRS)) {
     /* obj attributes */
-    res = hwloc__export_synthetic_obj_attr(topology, obj, tmp, tmplen);
+    res = hwloc__export_synthetic_obj_attr(topology, flags, obj, tmp, tmplen);
     if (hwloc__export_synthetic_update_status(&ret, &tmp, &tmplen, res) < 0)
       return -1;
   }
@@ -1339,7 +1405,7 @@ hwloc__export_synthetic_memory_children(struct hwloc_topology * topology, unsign
 
   if (flags & HWLOC_TOPOLOGY_EXPORT_SYNTHETIC_FLAG_V1) {
     /* v1: export a single NUMA child */
-    if (parent->memory_arity > 1 || mchild->type != HWLOC_OBJ_NUMANODE) {
+    if (parent->memory_arity > 1) {
       /* not supported */
       if (verbose)
 	fprintf(stderr, "Cannot export to synthetic v1 if multiple memory children are attached to the same location.\n");
@@ -1350,6 +1416,9 @@ hwloc__export_synthetic_memory_children(struct hwloc_topology * topology, unsign
     if (needprefix)
       hwloc__export_synthetic_add_char(&ret, &tmp, &tmplen, ' ');
 
+    /* ignore memcaches and export the NUMA node */
+    while (mchild->type != HWLOC_OBJ_NUMANODE)
+      mchild = mchild->memory_first_child;
     res = hwloc__export_synthetic_obj(topology, flags, mchild, 1, tmp, tmplen);
     if (hwloc__export_synthetic_update_status(&ret, &tmp, &tmplen, res) < 0)
       return -1;
@@ -1357,16 +1426,25 @@ hwloc__export_synthetic_memory_children(struct hwloc_topology * topology, unsign
   }
 
   while (mchild) {
-    /* FIXME: really recurse to export memcaches and numanode,
+    /* The core doesn't support shared memcache for now (because ACPI and Linux don't).
+     * So, for each mchild here, recurse only in the first children at each level.
+     *
+     * FIXME: whenever supported by the core, really recurse to export memcaches and numanode,
      * but it requires clever parsing of [ memcache [numa] [numa] ] during import,
      * better attaching of things to describe the hierarchy.
      */
     hwloc_obj_t numanode = mchild;
-    /* only export the first NUMA node leaf of each memory child
+    /* Only export the first NUMA node leaf of each memory child.
-     * FIXME: This assumes mscache aren't shared between nodes, that's true in current platforms
+     * Memcaches are ignored here, they will be summed and exported as a single attribute
+     * of the NUMA node in hwloc__export_synthetic_obj().
      */
     while (numanode && numanode->type != HWLOC_OBJ_NUMANODE) {
-      assert(numanode->arity == 1);
+      if (verbose && numanode->memory_arity > 1) {
+        static int warned = 0;
+        if (!warned)
+          fprintf(stderr, "Ignoring non-first memory children at non-first level of memory hierarchy.\n");
+        warned = 1;
+      }
       numanode = numanode->memory_first_child;
     }
     assert(numanode); /* there's always a numanode at the bottom of the memory tree */
@@ -1499,17 +1577,21 @@ hwloc_topology_export_synthetic(struct hwloc_topology * topology,
 
   if (flags & HWLOC_TOPOLOGY_EXPORT_SYNTHETIC_FLAG_V1) {
     /* v1 requires all NUMA at the same level */
-    hwloc_obj_t node;
+    hwloc_obj_t node, parent;
     signed pdepth;
 
     node = hwloc_get_obj_by_type(topology, HWLOC_OBJ_NUMANODE, 0);
     assert(node);
-    assert(hwloc__obj_type_is_normal(node->parent->type)); /* only depth-1 memory children for now */
+    parent = node->parent;
-    pdepth = node->parent->depth;
+    while (!hwloc__obj_type_is_normal(parent->type))
+      parent = parent->parent;
+    pdepth = parent->depth;
 
     while ((node = node->next_cousin) != NULL) {
-      assert(hwloc__obj_type_is_normal(node->parent->type)); /* only depth-1 memory children for now */
+      parent = node->parent;
-      if (node->parent->depth != pdepth) {
+      while (!hwloc__obj_type_is_normal(parent->type))
+        parent = parent->parent;
+      if (parent->depth != pdepth) {
 	if (verbose)
 	  fprintf(stderr, "Cannot export to synthetic v1 if memory is attached to parents at different depths.\n");
 	errno = EINVAL;
@@ -1522,7 +1604,7 @@ hwloc_topology_export_synthetic(struct hwloc_topology * topology,
 
   if (!(flags & HWLOC_TOPOLOGY_EXPORT_SYNTHETIC_FLAG_NO_ATTRS)) {
     /* obj attributes */
-    res = hwloc__export_synthetic_obj_attr(topology, obj, tmp, tmplen);
+    res = hwloc__export_synthetic_obj_attr(topology, flags, obj, tmp, tmplen);
     if (res > 0)
       needprefix = 1;
     if (hwloc__export_synthetic_update_status(&ret, &tmp, &tmplen, res) < 0)

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2025 Inria.  All rights reserved.
  * Copyright © 2009-2012, 2020 Université Bordeaux
  * Copyright © 2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -11,7 +11,9 @@
 
 #include "private/autogen/config.h"
 #include "hwloc.h"
+#include "hwloc/windows.h"
 #include "private/private.h"
+#include "private/windows.h" /* must be before windows.h */
 #include "private/debug.h"
 
 #include <windows.h>
@@ -54,6 +56,9 @@ typedef enum _LOGICAL_PROCESSOR_RELATIONSHIP {
   RelationCache,
   RelationProcessorPackage,
   RelationGroup,
+  RelationProcessorDie,
+  RelationNumaNodeEx, /* only used to *request* extended numa info only, but included in RelationAll, never returned on output */
+  RelationProcessorModule,
   RelationAll = 0xffff
 } LOGICAL_PROCESSOR_RELATIONSHIP;
 #else /* HAVE_LOGICAL_PROCESSOR_RELATIONSHIP */
@@ -62,28 +67,13 @@ typedef enum _LOGICAL_PROCESSOR_RELATIONSHIP {
 #    define RelationGroup 4
 #    define RelationAll 0xffff
 #  endif /* HAVE_RELATIONPROCESSORPACKAGE */
+#  ifndef HAVE_RELATIONPROCESSORDIE
+#    define RelationProcessorDie 5
+#    define RelationNumaNodeEx 6
+#    define RelationProcessorModule 7
+#  endif
 #endif /* HAVE_LOGICAL_PROCESSOR_RELATIONSHIP */
 
-#ifndef HAVE_SYSTEM_LOGICAL_PROCESSOR_INFORMATION
-typedef struct _SYSTEM_LOGICAL_PROCESSOR_INFORMATION {
-  ULONG_PTR ProcessorMask;
-  LOGICAL_PROCESSOR_RELATIONSHIP Relationship;
-  _ANONYMOUS_UNION
-  union {
-    struct {
-      BYTE flags;
-    } ProcessorCore;
-    struct {
-      DWORD NodeNumber;
-    } NumaNode;
-    CACHE_DESCRIPTOR Cache;
-    ULONGLONG Reserved[2];
-  } DUMMYUNIONNAME;
-} SYSTEM_LOGICAL_PROCESSOR_INFORMATION, *PSYSTEM_LOGICAL_PROCESSOR_INFORMATION;
-#endif
-
-/* Extended interface, for group support */
-
 #ifndef HAVE_GROUP_AFFINITY
 typedef struct _GROUP_AFFINITY {
   KAFFINITY Mask;
@@ -92,35 +82,40 @@ typedef struct _GROUP_AFFINITY {
 } GROUP_AFFINITY, *PGROUP_AFFINITY;
 #endif
 
-#ifndef HAVE_PROCESSOR_RELATIONSHIP
+/* always use our own structure because the EfficiencyClass field didn't exist before Win10 */
 typedef struct HWLOC_PROCESSOR_RELATIONSHIP {
   BYTE Flags;
-  BYTE EfficiencyClass; /* for RelationProcessorCore, higher means greater performance but less efficiency, only available in Win10+ */
+  BYTE EfficiencyClass; /* for RelationProcessorCore, higher means greater performance but less efficiency */
   BYTE Reserved[20];
   WORD GroupCount;
   GROUP_AFFINITY GroupMask[ANYSIZE_ARRAY];
-} PROCESSOR_RELATIONSHIP, *PPROCESSOR_RELATIONSHIP;
+} HWLOC_PROCESSOR_RELATIONSHIP;
-#endif
 
-#ifndef HAVE_NUMA_NODE_RELATIONSHIP
+/* always use our own structure because the GroupCount and GroupMasks fields didn't exist in some Win10 */
-typedef struct _NUMA_NODE_RELATIONSHIP {
+typedef struct HWLOC_NUMA_NODE_RELATIONSHIP {
   DWORD NodeNumber;
-  BYTE Reserved[20];
+  BYTE Reserved[18];
-  GROUP_AFFINITY GroupMask;
+  WORD GroupCount;
-} NUMA_NODE_RELATIONSHIP, *PNUMA_NODE_RELATIONSHIP;
+  _ANONYMOUS_UNION
-#endif
+  union {
+    GROUP_AFFINITY GroupMask;
+    GROUP_AFFINITY GroupMasks[ANYSIZE_ARRAY];
+  } DUMMYUNIONNAME;
+} HWLOC_NUMA_NODE_RELATIONSHIP;
 
-#ifndef HAVE_CACHE_RELATIONSHIP
+typedef struct HWLOC_CACHE_RELATIONSHIP {
-typedef struct _CACHE_RELATIONSHIP {
   BYTE Level;
   BYTE Associativity;
   WORD LineSize;
   DWORD CacheSize;
   PROCESSOR_CACHE_TYPE Type;
-  BYTE Reserved[20];
+  BYTE Reserved[18];
-  GROUP_AFFINITY GroupMask;
+  WORD GroupCount;
-} CACHE_RELATIONSHIP, *PCACHE_RELATIONSHIP;
+  union {
-#endif
+    GROUP_AFFINITY GroupMask;
+    GROUP_AFFINITY GroupMasks[ANYSIZE_ARRAY];
+  } DUMMYUNIONNAME;
+} HWLOC_CACHE_RELATIONSHIP;
 
 #ifndef HAVE_PROCESSOR_GROUP_INFO
 typedef struct _PROCESSOR_GROUP_INFO {
@@ -140,20 +135,19 @@ typedef struct _GROUP_RELATIONSHIP {
 } GROUP_RELATIONSHIP, *PGROUP_RELATIONSHIP;
 #endif
 
-#ifndef HAVE_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX
+/* always use our own structure because we need our own HWLOC_PROCESSOR/CACHE/NUMA_NODE_RELATIONSHIP */
-typedef struct _SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX {
+typedef struct HWLOC_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX {
   LOGICAL_PROCESSOR_RELATIONSHIP Relationship;
   DWORD Size;
   _ANONYMOUS_UNION
   union {
-    PROCESSOR_RELATIONSHIP Processor;
+    HWLOC_PROCESSOR_RELATIONSHIP Processor;
-    NUMA_NODE_RELATIONSHIP NumaNode;
+    HWLOC_NUMA_NODE_RELATIONSHIP NumaNode;
-    CACHE_RELATIONSHIP Cache;
+    HWLOC_CACHE_RELATIONSHIP Cache;
     GROUP_RELATIONSHIP Group;
     /* Odd: no member to tell the cpu mask of the package... */
   } DUMMYUNIONNAME;
-} SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX, *PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX;
+} HWLOC_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX;
-#endif
 
 #ifndef HAVE_PSAPI_WORKING_SET_EX_BLOCK
 typedef union _PSAPI_WORKING_SET_EX_BLOCK {
@@ -190,9 +184,6 @@ typedef struct _PROCESSOR_NUMBER {
 typedef WORD (WINAPI *PFN_GETACTIVEPROCESSORGROUPCOUNT)(void);
 static PFN_GETACTIVEPROCESSORGROUPCOUNT GetActiveProcessorGroupCountProc;
 
-static unsigned long nr_processor_groups = 1;
-static unsigned long max_numanode_index = 0;
-
 typedef WORD (WINAPI *PFN_GETACTIVEPROCESSORCOUNT)(WORD);
 static PFN_GETACTIVEPROCESSORCOUNT GetActiveProcessorCountProc;
 
@@ -202,10 +193,7 @@ static PFN_GETCURRENTPROCESSORNUMBER GetCurrentProcessorNumberProc;
 typedef VOID (WINAPI *PFN_GETCURRENTPROCESSORNUMBEREX)(PPROCESSOR_NUMBER);
 static PFN_GETCURRENTPROCESSORNUMBEREX GetCurrentProcessorNumberExProc;
 
-typedef BOOL (WINAPI *PFN_GETLOGICALPROCESSORINFORMATION)(PSYSTEM_LOGICAL_PROCESSOR_INFORMATION Buffer, PDWORD ReturnLength);
+typedef BOOL (WINAPI *PFN_GETLOGICALPROCESSORINFORMATIONEX)(LOGICAL_PROCESSOR_RELATIONSHIP relationship, HWLOC_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *Buffer, PDWORD ReturnLength);
-static PFN_GETLOGICALPROCESSORINFORMATION GetLogicalProcessorInformationProc;
-
-typedef BOOL (WINAPI *PFN_GETLOGICALPROCESSORINFORMATIONEX)(LOGICAL_PROCESSOR_RELATIONSHIP relationship, PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX Buffer, PDWORD ReturnLength);
 static PFN_GETLOGICALPROCESSORINFORMATIONEX GetLogicalProcessorInformationExProc;
 
 typedef BOOL (WINAPI *PFN_SETTHREADGROUPAFFINITY)(HANDLE hThread, const GROUP_AFFINITY *GroupAffinity, PGROUP_AFFINITY PreviousGroupAffinity);
@@ -240,14 +228,12 @@ static void hwloc_win_get_function_ptrs(void)
 #pragma GCC diagnostic ignored "-Wcast-function-type"
 #endif
 
-    kernel32 = LoadLibrary("kernel32.dll");
+    kernel32 = LoadLibrary(TEXT("kernel32.dll"));
     if (kernel32) {
       GetActiveProcessorGroupCountProc =
 	(PFN_GETACTIVEPROCESSORGROUPCOUNT) GetProcAddress(kernel32, "GetActiveProcessorGroupCount");
       GetActiveProcessorCountProc =
 	(PFN_GETACTIVEPROCESSORCOUNT) GetProcAddress(kernel32, "GetActiveProcessorCount");
-      GetLogicalProcessorInformationProc =
-	(PFN_GETLOGICALPROCESSORINFORMATION) GetProcAddress(kernel32, "GetLogicalProcessorInformation");
       GetCurrentProcessorNumberProc =
 	(PFN_GETCURRENTPROCESSORNUMBER) GetProcAddress(kernel32, "GetCurrentProcessorNumber");
       GetCurrentProcessorNumberExProc =
@@ -270,16 +256,13 @@ static void hwloc_win_get_function_ptrs(void)
 	(PFN_VIRTUALFREEEX) GetProcAddress(kernel32, "VirtualFreeEx");
     }
 
-    if (GetActiveProcessorGroupCountProc)
-      nr_processor_groups = GetActiveProcessorGroupCountProc();
-
     if (!QueryWorkingSetExProc) {
-      HMODULE psapi = LoadLibrary("psapi.dll");
+      HMODULE psapi = LoadLibrary(TEXT("psapi.dll"));
       if (psapi)
         QueryWorkingSetExProc = (PFN_QUERYWORKINGSETEX) GetProcAddress(psapi, "QueryWorkingSetEx");
     }
 
-    ntdll = GetModuleHandle("ntdll");
+    ntdll = GetModuleHandle(TEXT("ntdll"));
     RtlGetVersionProc = (PFN_RTLGETVERSION) GetProcAddress(ntdll, "RtlGetVersion");
 
 #if HWLOC_HAVE_GCC_W_CAST_FUNCTION_TYPE
@@ -363,6 +346,173 @@ static int hwloc_bitmap_to_single_ULONG_PTR(hwloc_const_bitmap_t set, unsigned *
   return 0;
 }
 
+/**********************
+ * Processor Groups
+ */
+
+static unsigned long max_numanode_index = 0;
+
+static unsigned long nr_processor_groups = 1;
+static hwloc_cpuset_t * processor_group_cpusets = NULL;
+
+static void
+hwloc_win_get_processor_groups(void)
+{
+  HWLOC_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *procInfoTotal, *tmpprocInfoTotal, *procInfo;
+  DWORD length;
+  unsigned i;
+
+  hwloc_debug("querying windows processor groups\n");
+
+  if (!GetLogicalProcessorInformationExProc)
+    goto error;
+
+  nr_processor_groups = GetActiveProcessorGroupCountProc();
+  if (!nr_processor_groups)
+    goto error;
+
+  hwloc_debug("found %lu windows processor groups\n", nr_processor_groups);
+
+  if (nr_processor_groups > 1 && SIZEOF_VOID_P == 4) {
+    if (HWLOC_SHOW_CRITICAL_ERRORS())
+      fprintf(stderr, "hwloc/windows: multiple processor groups found on 32bits Windows, topology may be invalid/incomplete.\n");
+  }
+
+  length = 0;
+  procInfoTotal = NULL;
+
+  while (1) {
+    if (GetLogicalProcessorInformationExProc(RelationGroup, procInfoTotal, &length))
+      break;
+    if (GetLastError() != ERROR_INSUFFICIENT_BUFFER)
+      goto error;
+    tmpprocInfoTotal = realloc(procInfoTotal, length);
+    if (!tmpprocInfoTotal)
+      goto error_with_procinfo;
+    procInfoTotal = tmpprocInfoTotal;
+  }
+
+  processor_group_cpusets = calloc(nr_processor_groups, sizeof(*processor_group_cpusets));
+  if (!processor_group_cpusets)
+    goto error_with_procinfo;
+
+  for (procInfo = procInfoTotal;
+       (void*) procInfo < (void*) ((uintptr_t) procInfoTotal + length);
+       procInfo = (void*) ((uintptr_t) procInfo + procInfo->Size)) {
+    unsigned id;
+
+    assert(procInfo->Relationship == RelationGroup);
+
+    hwloc_debug("Found %u active windows processor groups\n",
+                (unsigned) procInfo->Group.ActiveGroupCount);
+    for (id = 0; id < procInfo->Group.ActiveGroupCount; id++) {
+      KAFFINITY mask;
+      hwloc_bitmap_t set;
+
+      set = hwloc_bitmap_alloc();
+      if (!set)
+        goto error_with_cpusets;
+
+      mask = procInfo->Group.GroupInfo[id].ActiveProcessorMask;
+      hwloc_debug("group %u with %u cpus mask 0x%llx\n", id,
+                  (unsigned) procInfo->Group.GroupInfo[id].ActiveProcessorCount, (unsigned long long) mask);
+      /* KAFFINITY is ULONG_PTR */
+      hwloc_bitmap_set_ith_ULONG_PTR(set, id, mask);
+      /* FIXME: what if running 32bits on a 64bits windows with 64-processor groups?
+       * ULONG_PTR is 32bits, so half the group is invisible?
+       * maybe scale id to id*8/sizeof(ULONG_PTR) so that groups are 64-PU aligned?
+       */
+      hwloc_debug_2args_bitmap("group %u %d bitmap %s\n", id, procInfo->Group.GroupInfo[id].ActiveProcessorCount, set);
+      processor_group_cpusets[id] = set;
+    }
+  }
+
+  free(procInfoTotal);
+  return;
+
+ error_with_cpusets:
+  for(i=0; i<nr_processor_groups; i++) {
+    if (processor_group_cpusets[i])
+      hwloc_bitmap_free(processor_group_cpusets[i]);
+  }
+  free(processor_group_cpusets);
+  processor_group_cpusets = NULL;
+ error_with_procinfo:
+  free(procInfoTotal);
+ error:
+  /* on error set nr to 1 and keep cpusets NULL. We'll use the topology cpuset whenever needed */
+  nr_processor_groups = 1;
+}
+
+static void
+hwloc_win_free_processor_groups(void)
+{
+  unsigned i;
+  for(i=0; i<nr_processor_groups; i++) {
+    if (processor_group_cpusets[i])
+      hwloc_bitmap_free(processor_group_cpusets[i]);
+  }
+  free(processor_group_cpusets);
+  processor_group_cpusets = NULL;
+  nr_processor_groups = 1;
+}
+
+
+int
+hwloc_windows_get_nr_processor_groups(hwloc_topology_t topology, unsigned long flags)
+{
+  if (!topology->is_loaded || !topology->is_thissystem) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  if (flags) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  return nr_processor_groups;
+}
+
+int
+hwloc_windows_get_processor_group_cpuset(hwloc_topology_t topology, unsigned pg_index, hwloc_cpuset_t cpuset, unsigned long flags)
+{
+  if (!topology->is_loaded || !topology->is_thissystem) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  if (!cpuset) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  if (flags) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  if (pg_index >= nr_processor_groups) {
+    errno = ENOENT;
+    return -1;
+  }
+
+  if (!processor_group_cpusets) {
+    assert(nr_processor_groups == 1);
+    /* we found no processor groups, return the entire topology as a single one */
+    hwloc_bitmap_copy(cpuset, topology->levels[0][0]->cpuset);
+    return 0;
+  }
+
+  if (!processor_group_cpusets[pg_index]) {
+    errno = ENOENT;
+    return -1;
+  }
+
+  hwloc_bitmap_copy(cpuset, processor_group_cpusets[pg_index]);
+  return 0;
+}
+
 /**************************************************************
  * hwloc PU numbering with respect to Windows processor groups
  *
@@ -845,9 +995,15 @@ hwloc_look_windows(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
   OSVERSIONINFOEX osvi;
   char versionstr[20];
   char hostname[122] = "";
-  unsigned hostname_size = sizeof(hostname);
+#if !defined(__CYGWIN__)
+  DWORD hostname_size = sizeof(hostname);
+#else
+  size_t hostname_size = sizeof(hostname);
+#endif
   int has_efficiencyclass = 0;
   struct hwloc_win_efficiency_classes eclasses;
+  char *env = getenv("HWLOC_WINDOWS_PROCESSOR_GROUP_OBJS");
+  int keep_pgroup_objs = (env && atoi(env));
 
   assert(dstatus->phase == HWLOC_DISC_PHASE_CPU);
 
@@ -878,137 +1034,8 @@ hwloc_look_windows(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
 
   GetSystemInfo(&SystemInfo);
 
-  if (!GetLogicalProcessorInformationExProc && GetLogicalProcessorInformationProc) {
-      PSYSTEM_LOGICAL_PROCESSOR_INFORMATION procInfo, tmpprocInfo;
-      unsigned id;
-      unsigned i;
-      struct hwloc_obj *obj;
-      hwloc_obj_type_t type;
-
-      length = 0;
-      procInfo = NULL;
-
-      while (1) {
-	if (GetLogicalProcessorInformationProc(procInfo, &length))
-	  break;
-	if (GetLastError() != ERROR_INSUFFICIENT_BUFFER)
-	  return -1;
-	tmpprocInfo = realloc(procInfo, length);
-	if (!tmpprocInfo) {
-	  free(procInfo);
-	  goto out;
-	}
-	procInfo = tmpprocInfo;
-      }
-
-      assert(!length || procInfo);
-
-      for (i = 0; i < length / sizeof(*procInfo); i++) {
-
-        /* Ignore unknown caches */
-	if (procInfo->Relationship == RelationCache
-		&& procInfo->Cache.Type != CacheUnified
-		&& procInfo->Cache.Type != CacheData
-		&& procInfo->Cache.Type != CacheInstruction)
-	  continue;
-
-	id = HWLOC_UNKNOWN_INDEX;
-	switch (procInfo[i].Relationship) {
-	  case RelationNumaNode:
-	    type = HWLOC_OBJ_NUMANODE;
-	    id = procInfo[i].NumaNode.NodeNumber;
-	    gotnuma++;
-	    if (id > max_numanode_index)
-	      max_numanode_index = id;
-	    break;
-	  case RelationProcessorPackage:
-	    type = HWLOC_OBJ_PACKAGE;
-	    break;
-	  case RelationCache:
-	    type = (procInfo[i].Cache.Type == CacheInstruction ? HWLOC_OBJ_L1ICACHE : HWLOC_OBJ_L1CACHE) + procInfo[i].Cache.Level - 1;
-	    break;
-	  case RelationProcessorCore:
-	    type = HWLOC_OBJ_CORE;
-	    break;
-	  case RelationGroup:
-	  default:
-	    type = HWLOC_OBJ_GROUP;
-	    break;
-	}
-
-	if (!hwloc_filter_check_keep_object_type(topology, type))
-	  continue;
-
-	obj = hwloc_alloc_setup_object(topology, type, id);
-        obj->cpuset = hwloc_bitmap_alloc();
-	hwloc_debug("%s#%u mask %llx\n", hwloc_obj_type_string(type), id, (unsigned long long) procInfo[i].ProcessorMask);
-	/* ProcessorMask is a ULONG_PTR */
-	hwloc_bitmap_set_ith_ULONG_PTR(obj->cpuset, 0, procInfo[i].ProcessorMask);
-	hwloc_debug_2args_bitmap("%s#%u bitmap %s\n", hwloc_obj_type_string(type), id, obj->cpuset);
-
-	switch (type) {
-	  case HWLOC_OBJ_NUMANODE:
-	    {
-	      ULONGLONG avail;
-	      obj->nodeset = hwloc_bitmap_alloc();
-	      hwloc_bitmap_set(obj->nodeset, id);
-	      if ((GetNumaAvailableMemoryNodeExProc && GetNumaAvailableMemoryNodeExProc(id, &avail))
-		  || (GetNumaAvailableMemoryNodeProc && GetNumaAvailableMemoryNodeProc(id, &avail))) {
-		obj->attr->numanode.local_memory = avail;
-		gotnumamemory++;
-	      }
-	      obj->attr->numanode.page_types_len = 2;
-	      obj->attr->numanode.page_types = malloc(2 * sizeof(*obj->attr->numanode.page_types));
-	      memset(obj->attr->numanode.page_types, 0, 2 * sizeof(*obj->attr->numanode.page_types));
-	      obj->attr->numanode.page_types_len = 1;
-	      obj->attr->numanode.page_types[0].size = SystemInfo.dwPageSize;
-#if HAVE_DECL__SC_LARGE_PAGESIZE
-	      obj->attr->numanode.page_types_len++;
-	      obj->attr->numanode.page_types[1].size = sysconf(_SC_LARGE_PAGESIZE);
-#endif
-	      break;
-	    }
-	  case HWLOC_OBJ_L1CACHE:
-	  case HWLOC_OBJ_L2CACHE:
-	  case HWLOC_OBJ_L3CACHE:
-	  case HWLOC_OBJ_L4CACHE:
-	  case HWLOC_OBJ_L5CACHE:
-	  case HWLOC_OBJ_L1ICACHE:
-	  case HWLOC_OBJ_L2ICACHE:
-	  case HWLOC_OBJ_L3ICACHE:
-	    obj->attr->cache.size = procInfo[i].Cache.Size;
-	    obj->attr->cache.associativity = procInfo[i].Cache.Associativity == CACHE_FULLY_ASSOCIATIVE ? -1 : procInfo[i].Cache.Associativity ;
-	    obj->attr->cache.linesize = procInfo[i].Cache.LineSize;
-	    obj->attr->cache.depth = procInfo[i].Cache.Level;
-	    switch (procInfo->Cache.Type) {
-	      case CacheUnified:
-		obj->attr->cache.type = HWLOC_OBJ_CACHE_UNIFIED;
-		break;
-	      case CacheData:
-		obj->attr->cache.type = HWLOC_OBJ_CACHE_DATA;
-		break;
-	      case CacheInstruction:
-		obj->attr->cache.type = HWLOC_OBJ_CACHE_INSTRUCTION;
-		break;
-	      default:
-		hwloc_free_unlinked_object(obj);
-		continue;
-	    }
-	    break;
-	  case HWLOC_OBJ_GROUP:
-	    obj->attr->group.kind = procInfo[i].Relationship == RelationGroup ? HWLOC_GROUP_KIND_WINDOWS_PROCESSOR_GROUP : HWLOC_GROUP_KIND_WINDOWS_RELATIONSHIP_UNKNOWN;
-	    break;
-	  default:
-	    break;
-	}
-	hwloc__insert_object_by_cpuset(topology, NULL, obj, "windows:GetLogicalProcessorInformation");
-      }
-
-      free(procInfo);
-  }
-
   if (GetLogicalProcessorInformationExProc) {
-      PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX procInfoTotal, tmpprocInfoTotal, procInfo;
+      HWLOC_SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *procInfoTotal, *tmpprocInfoTotal, *procInfo;
       unsigned id;
       struct hwloc_obj *obj;
       hwloc_obj_type_t type;
@@ -1036,19 +1063,32 @@ hwloc_look_windows(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
         unsigned efficiency_class = 0;
         GROUP_AFFINITY *GroupMask;
 
-        /* Ignore unknown caches */
+	if (procInfo->Relationship == RelationCache) {
-	if (procInfo->Relationship == RelationCache
+          if (!topology->want_some_cpu_caches)
-		&& procInfo->Cache.Type != CacheUnified
+            /* TODO: check if RelationAll&~RelationCache works? */
-		&& procInfo->Cache.Type != CacheData
+            continue;
-		&& procInfo->Cache.Type != CacheInstruction)
+          if (procInfo->Cache.Type != CacheUnified
-	  continue;
+              && procInfo->Cache.Type != CacheData
+              && procInfo->Cache.Type != CacheInstruction)
+            /* Ignore unknown caches */
+            continue;
+        }
 
 	id = HWLOC_UNKNOWN_INDEX;
 	switch (procInfo->Relationship) {
+          case RelationNumaNodeEx: /* only used on input anyway */
 	  case RelationNumaNode:
 	    type = HWLOC_OBJ_NUMANODE;
-            num = 1;
+            /* Starting with Windows 11 and Server 2022, the GroupCount field is valid and >=1
-            GroupMask = &procInfo->NumaNode.GroupMask;
+             * and we may read GroupMasks[]. Older releases have GroupCount==0 and we must read GroupMask.
+             */
+            if (procInfo->NumaNode.GroupCount) {
+              num = procInfo->NumaNode.GroupCount;
+              GroupMask = procInfo->NumaNode.GroupMasks;
+            } else {
+              num = 1;
+              GroupMask = &procInfo->NumaNode.GroupMask;
+            }
 	    id = procInfo->NumaNode.NodeNumber;
 	    gotnuma++;
 	    if (id > max_numanode_index)
@@ -1056,23 +1096,35 @@ hwloc_look_windows(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
 	    break;
 	  case RelationProcessorPackage:
 	    type = HWLOC_OBJ_PACKAGE;
+	    num = procInfo->Processor.GroupCount;
+	    GroupMask = procInfo->Processor.GroupMask;
+	    break;
+	  case RelationProcessorDie:
+            type = HWLOC_OBJ_DIE;
             num = procInfo->Processor.GroupCount;
             GroupMask = procInfo->Processor.GroupMask;
-	    break;
+            break;
+	  case RelationProcessorModule:
+            type = HWLOC_OBJ_GROUP;
+            num = procInfo->Processor.GroupCount;
+            GroupMask = procInfo->Processor.GroupMask;
+            break;
 	  case RelationCache:
 	    type = (procInfo->Cache.Type == CacheInstruction ? HWLOC_OBJ_L1ICACHE : HWLOC_OBJ_L1CACHE) + procInfo->Cache.Level - 1;
-            num = 1;
+            /* GroupCount added approximately with NumaNode.GroupCount above */
-            GroupMask = &procInfo->Cache.GroupMask;
+            if (procInfo->Cache.GroupCount) {
+              num = procInfo->Cache.GroupCount;
+              GroupMask = procInfo->Cache.GroupMasks;
+            } else {
+              num = 1;
+              GroupMask = &procInfo->Cache.GroupMask;
+            }
 	    break;
 	  case RelationProcessorCore:
 	    type = HWLOC_OBJ_CORE;
             num = procInfo->Processor.GroupCount;
             GroupMask = procInfo->Processor.GroupMask;
-            if (has_efficiencyclass)
+            efficiency_class = procInfo->Processor.EfficiencyClass;
-              /* the EfficiencyClass field didn't exist before Windows10 and recent MSVC headers,
-               * so just access it manually instead of trying to detect it.
-               */
-              efficiency_class = * ((&procInfo->Processor.Flags) + 1);
 	    break;
 	  case RelationGroup:
 	    /* So strange an interface... */
@@ -1097,11 +1149,12 @@ hwloc_look_windows(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
 		groups_pu_set = hwloc_bitmap_alloc();
 	      hwloc_bitmap_or(groups_pu_set, groups_pu_set, set);
 
-	      if (hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_GROUP)) {
+              /* Ignore processor groups unless requested and filtered-in */
+              if (keep_pgroup_objs && hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_GROUP)) {
 		obj = hwloc_alloc_setup_object(topology, HWLOC_OBJ_GROUP, id);
 		obj->cpuset = set;
 		obj->attr->group.kind = HWLOC_GROUP_KIND_WINDOWS_PROCESSOR_GROUP;
-		hwloc__insert_object_by_cpuset(topology, NULL, obj, "windows:GetLogicalProcessorInformation:ProcessorGroup");
+		hwloc__insert_object_by_cpuset(topology, NULL, obj, "windows:GetLogicalProcessorInformationEx:ProcessorGroup");
 	      } else
 		hwloc_bitmap_free(set);
 	    }
@@ -1177,6 +1230,19 @@ hwloc_look_windows(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
 		continue;
 	    }
 	    break;
+          case HWLOC_OBJ_GROUP:
+            switch (procInfo->Relationship) {
+            case RelationGroup:
+              obj->attr->group.kind = HWLOC_GROUP_KIND_WINDOWS_PROCESSOR_GROUP;
+              break;
+            case RelationProcessorModule:
+              obj->attr->group.kind = HWLOC_GROUP_KIND_INTEL_MODULE;
+              obj->subtype = strdup("Module");
+              break;
+            default:
+              obj->attr->group.kind = HWLOC_GROUP_KIND_WINDOWS_RELATIONSHIP_UNKNOWN;
+            }
+            break;
 	  default:
 	    break;
 	}
@@ -1328,11 +1394,13 @@ hwloc_set_windows_hooks(struct hwloc_binding_hooks *hooks,
 static int hwloc_windows_component_init(unsigned long flags __hwloc_attribute_unused)
 {
   hwloc_win_get_function_ptrs();
+  hwloc_win_get_processor_groups();
   return 0;
 }
 
 static void hwloc_windows_component_finalize(unsigned long flags __hwloc_attribute_unused)
 {
+  hwloc_win_free_processor_groups();
 }
 
 static struct hwloc_backend *

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

@@ -1,17 +1,21 @@
 /*
- * Copyright © 2010-2020 Inria.  All rights reserved.
+ * Copyright © 2010-2025 Inria.  All rights reserved.
  * Copyright © 2010-2013 Université Bordeaux
  * Copyright © 2010-2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
  *
  *
- * This backend is only used when the operating system does not export
+ * This backend is mostly used when the operating system does not export
  * the necessary hardware topology information to user-space applications.
- * Currently, only the FreeBSD backend relies on this x86 backend.
+ * Currently, FreeBSD and NetBSD only add PUs and then fallback to this
+ * backend for CPU/Cache discovery.
  *
  * Other backends such as Linux have their own way to retrieve various
  * pieces of hardware topology information from the operating system
  * on various architectures, without having to use this x86-specific code.
+ * But this backend is still used after them to annotate some objects with
+ * additional details (CPU info in Package, Inclusiveness in Caches).
+ * It may also be enabled manually to work-around bugs in native OS discovery.
  */
 
 #include "private/autogen/config.h"
@@ -35,6 +39,12 @@ struct hwloc_x86_backend_data_s {
   int apicid_unique;
   char *src_cpuiddump_path;
   int is_knl;
+  int is_hybrid;
+  int found_die_ids;
+  int found_complex_ids;
+  int found_unit_ids;
+  int found_module_ids;
+  int found_tile_ids;
 };
 
 /************************************
@@ -77,7 +87,7 @@ cpuiddump_read(const char *dirpath, unsigned idx)
 
   cpuiddump = malloc(sizeof(*cpuiddump));
   if (!cpuiddump) {
-    fprintf(stderr, "Failed to allocate cpuiddump for PU #%u, ignoring cpuiddump.\n", idx);
+    fprintf(stderr, "hwloc/x86: Failed to allocate cpuiddump for PU #%u, ignoring cpuiddump.\n", idx);
     goto out;
   }
 
@@ -88,7 +98,7 @@ cpuiddump_read(const char *dirpath, unsigned idx)
   snprintf(filename, filenamelen, "%s/pu%u", dirpath, idx);
   file = fopen(filename, "r");
   if (!file) {
-    fprintf(stderr, "Could not read dumped cpuid file %s, ignoring cpuiddump.\n", filename);
+    fprintf(stderr, "hwloc/x86: Could not read dumped cpuid file %s, ignoring cpuiddump.\n", filename);
     goto out_with_filename;
   }
 
@@ -97,7 +107,7 @@ cpuiddump_read(const char *dirpath, unsigned idx)
     nr++;
   cpuiddump->entries = malloc(nr * sizeof(struct cpuiddump_entry));
   if (!cpuiddump->entries) {
-    fprintf(stderr, "Failed to allocate %u cpuiddump entries for PU #%u, ignoring cpuiddump.\n", nr, idx);
+    fprintf(stderr, "hwloc/x86: Failed to allocate %u cpuiddump entries for PU #%u, ignoring cpuiddump.\n", nr, idx);
     goto out_with_file;
   }
 
@@ -153,7 +163,7 @@ cpuiddump_find_by_input(unsigned *eax, unsigned *ebx, unsigned *ecx, unsigned *e
     return;
   }
 
-  fprintf(stderr, "Couldn't find %x,%x,%x,%x in dumped cpuid, returning 0s.\n",
+  fprintf(stderr, "hwloc/x86: Couldn't find %x,%x,%x,%x in dumped cpuid, returning 0s.\n",
 	  *eax, *ebx, *ecx, *edx);
   *eax = 0;
   *ebx = 0;
@@ -207,7 +217,8 @@ struct procinfo {
 #define TILE 4
 #define MODULE 5
 #define DIE 6
-#define HWLOC_X86_PROCINFO_ID_NR 7
+#define COMPLEX 7
+#define HWLOC_X86_PROCINFO_ID_NR 8
   unsigned ids[HWLOC_X86_PROCINFO_ID_NR];
   unsigned *otherids;
   unsigned levels;
@@ -311,7 +322,7 @@ static void read_amd_caches_topoext(struct procinfo *infos, struct cpuiddump *sr
   /* the code below doesn't want any other cache yet */
   assert(!infos->numcaches);
 
-  for (cachenum = 0; ; cachenum++) {
+  for (cachenum = 0; cachenum<16 /* guard */; cachenum++) {
     eax = 0x8000001d;
     ecx = cachenum;
     cpuid_or_from_dump(&eax, &ebx, &ecx, &edx, src_cpuiddump);
@@ -322,7 +333,7 @@ static void read_amd_caches_topoext(struct procinfo *infos, struct cpuiddump *sr
 
   cache = infos->cache = malloc(infos->numcaches * sizeof(*infos->cache));
   if (cache) {
-    for (cachenum = 0; ; cachenum++) {
+    for (cachenum = 0; cachenum<16 /* guard */; cachenum++) {
       unsigned long linesize, linepart, ways, sets;
       eax = 0x8000001d;
       ecx = cachenum;
@@ -375,7 +386,7 @@ static void read_intel_caches(struct hwloc_x86_backend_data_s *data, struct proc
   unsigned cachenum;
   struct cacheinfo *cache;
 
-  for (cachenum = 0; ; cachenum++) {
+  for (cachenum = 0; cachenum<16 /* guard */; cachenum++) {
     eax = 0x04;
     ecx = cachenum;
     cpuid_or_from_dump(&eax, &ebx, &ecx, &edx, src_cpuiddump);
@@ -397,7 +408,7 @@ static void read_intel_caches(struct hwloc_x86_backend_data_s *data, struct proc
     infos->cache = tmpcaches;
     cache = &infos->cache[oldnumcaches];
 
-    for (cachenum = 0; ; cachenum++) {
+    for (cachenum = 0; cachenum<16 /* guard */; cachenum++) {
       unsigned long linesize, linepart, ways, sets;
       eax = 0x04;
       ecx = cachenum;
@@ -477,7 +488,7 @@ static void read_amd_cores_legacy(struct procinfo *infos, struct cpuiddump *src_
 }
 
 /* AMD unit/node from CPUID 0x8000001e leaf (topoext) */
-static void read_amd_cores_topoext(struct procinfo *infos, unsigned long flags, struct cpuiddump *src_cpuiddump)
+static void read_amd_cores_topoext(struct hwloc_x86_backend_data_s *data, struct procinfo *infos, unsigned long flags __hwloc_attribute_unused, struct cpuiddump *src_cpuiddump)
 {
   unsigned apic_id, nodes_per_proc = 0;
   unsigned eax, ebx, ecx, edx;
@@ -486,7 +497,6 @@ static void read_amd_cores_topoext(struct procinfo *infos, unsigned long flags,
   cpuid_or_from_dump(&eax, &ebx, &ecx, &edx, src_cpuiddump);
   infos->apicid = apic_id = eax;
 
-  if (flags & HWLOC_X86_DISC_FLAG_TOPOEXT_NUMANODES) {
     if (infos->cpufamilynumber == 0x16) {
       /* ecx is reserved */
       infos->ids[NODE] = 0;
@@ -497,15 +507,16 @@ static void read_amd_cores_topoext(struct procinfo *infos, unsigned long flags,
       nodes_per_proc = ((ecx >> 8) & 7) + 1;
     }
     if ((infos->cpufamilynumber == 0x15 && nodes_per_proc > 2)
-	|| ((infos->cpufamilynumber == 0x17 || infos->cpufamilynumber == 0x18) && nodes_per_proc > 4)) {
+	|| ((infos->cpufamilynumber == 0x17 || infos->cpufamilynumber == 0x18) && nodes_per_proc > 4)
+        || (infos->cpufamilynumber == 0x19 && nodes_per_proc > 1)) {
       hwloc_debug("warning: undefined nodes_per_proc value %u, assuming it means %u\n", nodes_per_proc, nodes_per_proc);
     }
-  }
 
   if (infos->cpufamilynumber <= 0x16) { /* topoext appeared in 0x15 and compute-units were only used in 0x15 and 0x16 */
     unsigned cores_per_unit;
     /* coreid was obtained from read_amd_cores_legacy() earlier */
     infos->ids[UNIT] = ebx & 0xff;
+    data->found_unit_ids = 1;
     cores_per_unit = ((ebx >> 8) & 0xff) + 1;
     hwloc_debug("topoext %08x, %u nodes, node %u, %u cores in unit %u\n", apic_id, nodes_per_proc, infos->ids[NODE], cores_per_unit, infos->ids[UNIT]);
     /* coreid and unitid are package-wide (core 0-15 and unit 0-7 on 16-core 2-NUMAnode processor).
@@ -520,19 +531,29 @@ static void read_amd_cores_topoext(struct procinfo *infos, unsigned long flags,
   }
 }
 
-/* Intel core/thread or even die/module/tile from CPUID 0x0b or 0x1f leaves (v1 and v2 extended topology enumeration) */
+/* Intel core/thread or even die/module/tile from CPUID 0x0b or 0x1f leaves (v1 and v2 extended topology enumeration)
-static void read_intel_cores_exttopoenum(struct procinfo *infos, unsigned leaf, struct cpuiddump *src_cpuiddump)
+ * or AMD core/thread or even complex/ccd from CPUID 0x0b or 0x80000026 (extended CPU topology)
+ */
+static void read_extended_topo(struct hwloc_x86_backend_data_s *data, struct procinfo *infos, unsigned leaf, enum cpuid_type cpuid_type __hwloc_attribute_unused, struct cpuiddump *src_cpuiddump)
 {
-  unsigned level, apic_nextshift, apic_number, apic_type, apic_id = 0, apic_shift = 0, id;
+  unsigned level, apic_nextshift, apic_type, apic_id = 0, apic_shift = 0, id;
   unsigned threadid __hwloc_attribute_unused = 0; /* shut-up compiler */
   unsigned eax, ebx, ecx = 0, edx;
   int apic_packageshift = 0;
 
-  for (level = 0; ; level++) {
+  for (level = 0; level<32 /* guard */; level++) {
     ecx = level;
     eax = leaf;
     cpuid_or_from_dump(&eax, &ebx, &ecx, &edx, src_cpuiddump);
-    if (!eax && !ebx)
+    /* Intel specifies that the 0x0b/0x1f loop should stop when we get "invalid domain" (0 in ecx[8:15])
+     * (if so, we also get 0 in eax/ebx for invalid subleaves). Zhaoxin implements this too.
+     * However AMD rather says that the 0x80000026/0x0b loop should stop when we get "no thread at this level" (0 in ebx[0:15]).
+     *
+     * Linux kernel <= 6.8 used "invalid domain" for both Intel and AMD (in detect_extended_topology())
+     * but x86 discovery revamp in 6.9 now properly checks both Intel and AMD conditions (in topo_subleaf()).
+     * So let's assume we are allowed to break-out once one of the Intel+AMD conditions is met.
+     */
+    if (!(ebx & 0xffff) || !(ecx & 0xff00))
       break;
     apic_packageshift = eax & 0x1f;
   }
@@ -541,47 +562,68 @@ static void read_intel_cores_exttopoenum(struct procinfo *infos, unsigned leaf,
     infos->otherids = malloc(level * sizeof(*infos->otherids));
     if (infos->otherids) {
       infos->levels = level;
-      for (level = 0; ; level++) {
+      for (level = 0; level<32 /* guard */; level++) {
 	ecx = level;
 	eax = leaf;
 	cpuid_or_from_dump(&eax, &ebx, &ecx, &edx, src_cpuiddump);
-	if (!eax && !ebx)
+        if (!(ebx & 0xffff) || !(ecx & 0xff00))
-	  break;
+          break;
 	apic_nextshift = eax & 0x1f;
-	apic_number = ebx & 0xffff;
 	apic_type = (ecx & 0xff00) >> 8;
 	apic_id = edx;
 	id = (apic_id >> apic_shift) & ((1 << (apic_packageshift - apic_shift)) - 1);
-	hwloc_debug("x2APIC %08x %u: nextshift %u num %2u type %u id %2u\n", apic_id, level, apic_nextshift, apic_number, apic_type, id);
+	hwloc_debug("x2APIC %08x %u: nextshift %u nextnumber %2u type %u id %2u\n",
+                    apic_id,
+                    level,
+                    apic_nextshift,
+                    ebx & 0xffff /* number of threads in next level */,
+                    apic_type,
+                    id);
 	infos->apicid = apic_id;
 	infos->otherids[level] = UINT_MAX;
-	switch (apic_type) {
+        switch (apic_type) {
-	case 1:
+        case 1:
-	  threadid = id;
+          threadid = id;
-	  /* apic_number is the actual number of threads per core */
+          break;
-	  break;
+        case 2:
-	case 2:
+          infos->ids[CORE] = id;
-	  infos->ids[CORE] = id;
+          break;
-	  /* apic_number is the actual number of threads per die */
+        case 3:
-	  break;
+          if (leaf == 0x80000026) {
-	case 3:
+            data->found_complex_ids = 1;
-	  infos->ids[MODULE] = id;
+            infos->ids[COMPLEX] = id;
-	  /* apic_number is the actual number of threads per tile */
+          } else {
-	  break;
+            data->found_module_ids = 1;
-	case 4:
+            infos->ids[MODULE] = id;
-	  infos->ids[TILE] = id;
+          }
-	  /* apic_number is the actual number of threads per die */
+          break;
-	  break;
+        case 4:
-	case 5:
+          if (leaf == 0x80000026) {
-	  infos->ids[DIE] = id;
+            data->found_die_ids = 1;
-	  /* apic_number is the actual number of threads per package */
+            infos->ids[DIE] = id;
-	  break;
+          } else {
-	default:
+            data->found_tile_ids = 1;
-	  hwloc_debug("x2APIC %u: unknown type %u\n", level, apic_type);
+            infos->ids[TILE] = id;
-	  infos->otherids[level] = apic_id >> apic_shift;
+          }
-	  break;
+          break;
-	}
+        case 5:
-	apic_shift = apic_nextshift;
+          if (leaf == 0x80000026) {
+            goto unknown_type;
+          } else {
+            data->found_die_ids = 1;
+            infos->ids[DIE] = id;
+          }
+          break;
+        case 6:
+          /* TODO: "DieGrp" on Intel */
+          /* fallthrough */
+        default:
+        unknown_type:
+          hwloc_debug("x2APIC %u: unknown type %u\n", level, apic_type);
+          infos->otherids[level] = apic_id >> apic_shift;
+          break;
+        }
+        apic_shift = apic_nextshift;
       }
       infos->apicid = apic_id;
       infos->ids[PKG] = apic_id >> apic_shift;
@@ -610,10 +652,19 @@ static void look_proc(struct hwloc_backend *backend, struct procinfo *infos, uns
   eax = 0x01;
   cpuid_or_from_dump(&eax, &ebx, &ecx, &edx, src_cpuiddump);
   infos->apicid = ebx >> 24;
-  if (edx & (1 << 28))
+  if (edx & (1 << 28)) {
-    legacy_max_log_proc = 1 << hwloc_flsl(((ebx >> 16) & 0xff) - 1);
+    unsigned ebx_16_23 = (ebx >> 16) & 0xff;
-  else
+    if (ebx_16_23) {
+      legacy_max_log_proc = 1 << hwloc_flsl(ebx_16_23 - 1);
+    } else {
+      hwloc_debug("HTT bit set in CPUID 0x01.edx, but legacy_max_proc = 0 in ebx, assuming legacy_max_log_proc = 1\n");
+      legacy_max_log_proc = 1;
+    }
+  } else {
+    hwloc_debug("HTT bit not set in CPUID 0x01.edx, assuming legacy_max_log_proc = 1\n");
     legacy_max_log_proc = 1;
+  }
+
   hwloc_debug("APIC ID 0x%02x legacy_max_log_proc %u\n", infos->apicid, legacy_max_log_proc);
   infos->ids[PKG] = infos->apicid / legacy_max_log_proc;
   legacy_log_proc_id = infos->apicid % legacy_max_log_proc;
@@ -676,22 +727,34 @@ static void look_proc(struct hwloc_backend *backend, struct procinfo *infos, uns
       unsigned max_nbcores;
       unsigned max_nbthreads;
       unsigned threadid __hwloc_attribute_unused;
+      hwloc_debug("Trying to get core/thread IDs from 0x04...\n");
       max_nbcores = ((eax >> 26) & 0x3f) + 1;
-      max_nbthreads = legacy_max_log_proc / max_nbcores;
+      hwloc_debug("found %u cores max\n", max_nbcores);
-      hwloc_debug("thus %u threads\n", max_nbthreads);
+      /* some VMs (e.g. issue#525) don't report valid information, check things before dividing by 0. */
-      threadid = legacy_log_proc_id % max_nbthreads;
+      if (!max_nbcores) {
-      infos->ids[CORE] = legacy_log_proc_id / max_nbthreads;
+        hwloc_debug("cannot detect core/thread IDs from 0x04 without a valid max of cores\n");
-      hwloc_debug("this is thread %u of core %u\n", threadid, infos->ids[CORE]);
+      } else {
+        max_nbthreads = legacy_max_log_proc / max_nbcores;
+        hwloc_debug("found %u threads max\n", max_nbthreads);
+        if (!max_nbthreads) {
+          hwloc_debug("cannot detect core/thread IDs from 0x04 without a valid max of threads\n");
+        } else {
+          threadid = legacy_log_proc_id % max_nbthreads;
+          infos->ids[CORE] = legacy_log_proc_id / max_nbthreads;
+          hwloc_debug("this is thread %u of core %u\n", threadid, infos->ids[CORE]);
+        }
+      }
     }
   }
 
   if (highest_cpuid >= 0x1a && has_hybrid(features)) {
-    /* Get hybrid cpu information from cpuid 0x1a */
+    /* Get hybrid cpu information from cpuid 0x1a on Intel */
     eax = 0x1a;
     ecx = 0;
     cpuid_or_from_dump(&eax, &ebx, &ecx, &edx, src_cpuiddump);
     infos->hybridcoretype = eax >> 24;
     infos->hybridnativemodel = eax & 0xffffff;
+    data->is_hybrid = 1;
   }
 
   /*********************************************************************************
@@ -713,23 +776,30 @@ static void look_proc(struct hwloc_backend *backend, struct procinfo *infos, uns
      *
      * Only needed when x2apic supported if NUMA nodes are needed.
      */
-    read_amd_cores_topoext(infos, flags, src_cpuiddump);
+    read_amd_cores_topoext(data, infos, flags, src_cpuiddump);
   }
 
-  if ((cpuid_type == intel) && highest_cpuid >= 0x1f) {
+  if ((cpuid_type == amd) && highest_ext_cpuid >= 0x80000026) {
+    /* Get socket/die/complex/core/thread information from cpuid 0x80000026
+     * (AMD Extended CPU Topology)
+     */
+    read_extended_topo(data, infos, 0x80000026, cpuid_type, src_cpuiddump);
+
+  } else if ((cpuid_type == intel || cpuid_type == zhaoxin) && highest_cpuid >= 0x1f) {
     /* Get package/die/module/tile/core/thread information from cpuid 0x1f
      * (Intel v2 Extended Topology Enumeration)
      */
-    read_intel_cores_exttopoenum(infos, 0x1f, src_cpuiddump);
+    read_extended_topo(data, infos, 0x1f, cpuid_type, src_cpuiddump);
 
   } else if ((cpuid_type == intel || cpuid_type == amd || cpuid_type == zhaoxin)
 	     && highest_cpuid >= 0x0b && has_x2apic(features)) {
     /* Get package/core/thread information from cpuid 0x0b
      * (Intel v1 Extended Topology Enumeration)
      */
-    read_intel_cores_exttopoenum(infos, 0x0b, src_cpuiddump);
+    read_extended_topo(data, infos, 0x0b, cpuid_type, src_cpuiddump);
   }
 
+ if (backend->topology->want_some_cpu_caches) {
   /**************************************
    * Get caches from CPU-specific leaves
    */
@@ -772,13 +842,19 @@ static void look_proc(struct hwloc_backend *backend, struct procinfo *infos, uns
 
     } else if (cpuid_type == amd) {
       /* AMD quirks */
-      if (infos->cpufamilynumber == 0x17
+      if (infos->cpufamilynumber >= 0x17 && cache->level == 3) {
-	  && cache->level == 3 && cache->nbthreads_sharing == 6) {
+	/* AMD family 0x19 always shares L3 between 16 APIC ids (8 HT cores).
-	/* AMD family 0x17 always shares L3 between 8 APIC ids,
+         * while Family 0x17 shares between 8 APIC ids (4 HT cores).
-	 * even when only 6 APIC ids are enabled and reported in nbthreads_sharing
+         * But many models have less APIC ids enabled and reported in nbthreads_sharing.
-	 * (on 24-core CPUs).
+         * It means we must round-up nbthreads_sharing to the nearest power of 2
+         * before computing cacheid.
 	 */
-	cache->cacheid = infos->apicid / 8;
+        unsigned nbapics_sharing = cache->nbthreads_sharing;
+        if (nbapics_sharing & (nbapics_sharing-1))
+          /* not a power of two, round-up */
+          nbapics_sharing = 1U<<(1+hwloc_ffsl(nbapics_sharing));
+
+	cache->cacheid = infos->apicid / nbapics_sharing;
 
       } else if (infos->cpufamilynumber== 0x10 && infos->cpumodelnumber == 0x9
 	  && cache->level == 3
@@ -804,7 +880,7 @@ static void look_proc(struct hwloc_backend *backend, struct procinfo *infos, uns
       } else if (infos->cpufamilynumber == 0x15
 		 && (infos->cpumodelnumber == 0x1 /* Bulldozer */ || infos->cpumodelnumber == 0x2 /* Piledriver */)
 		 && cache->level == 3 && cache->nbthreads_sharing == 6) {
-	/* AMD Bulldozer and Piledriver 12-core processors have same APIC ids as Magny-Cours below,
+	/* AMD Bulldozer and Piledriver 12-core processors have same APIC ids as Magny-Cours above,
 	 * but we can't merge the checks because the original nbthreads_sharing must be exactly 6 here.
 	 */
 	cache->cacheid = (infos->apicid % legacy_max_log_proc) / cache->nbthreads_sharing /* cacheid within the package */
@@ -821,6 +897,7 @@ static void look_proc(struct hwloc_backend *backend, struct procinfo *infos, uns
       }
     }
   }
+ }
 
   if (hwloc_bitmap_isset(data->apicid_set, infos->apicid))
     data->apicid_unique = 0;
@@ -908,6 +985,16 @@ static void summarize(struct hwloc_backend *backend, struct procinfo *infos, uns
   int gotnuma = 0;
   int fulldiscovery = (flags & HWLOC_X86_DISC_FLAG_FULL);
 
+#ifdef HWLOC_DEBUG
+  hwloc_debug("\nSummary of x86 CPUID topology:\n");
+  for(i=0; i<nbprocs; i++) {
+    hwloc_debug("PU %u present=%u apicid=%u on PKG %d CORE %d DIE %d NODE %d\n",
+                i, infos[i].present, infos[i].apicid,
+                infos[i].ids[PKG], infos[i].ids[CORE], infos[i].ids[DIE], infos[i].ids[NODE]);
+  }
+  hwloc_debug("\n");
+#endif
+
   for (i = 0; i < nbprocs; i++)
     if (infos[i].present) {
       hwloc_bitmap_set(complete_cpuset, i);
@@ -1012,21 +1099,34 @@ static void summarize(struct hwloc_backend *backend, struct procinfo *infos, uns
 
   if (hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_GROUP)) {
     if (fulldiscovery) {
-      /* Look for AMD Compute units inside packages */
+      if (data->found_unit_ids) {
-      hwloc_bitmap_copy(remaining_cpuset, complete_cpuset);
+        /* Look for AMD Complex inside packages */
-      hwloc_x86_add_groups(topology, infos, nbprocs, remaining_cpuset,
+        hwloc_bitmap_copy(remaining_cpuset, complete_cpuset);
-			   UNIT, "Compute Unit",
+        hwloc_x86_add_groups(topology, infos, nbprocs, remaining_cpuset,
-			   HWLOC_GROUP_KIND_AMD_COMPUTE_UNIT, 0);
+                             COMPLEX, "Complex",
-      /* Look for Intel Modules inside packages */
+                             HWLOC_GROUP_KIND_AMD_COMPLEX, 0);
-      hwloc_bitmap_copy(remaining_cpuset, complete_cpuset);
+      }
-      hwloc_x86_add_groups(topology, infos, nbprocs, remaining_cpuset,
+      if (data->found_unit_ids) {
-			   MODULE, "Module",
+        /* Look for AMD Compute units inside packages */
-			   HWLOC_GROUP_KIND_INTEL_MODULE, 0);
+        hwloc_bitmap_copy(remaining_cpuset, complete_cpuset);
-      /* Look for Intel Tiles inside packages */
+        hwloc_x86_add_groups(topology, infos, nbprocs, remaining_cpuset,
-      hwloc_bitmap_copy(remaining_cpuset, complete_cpuset);
+                             UNIT, "Compute Unit",
-      hwloc_x86_add_groups(topology, infos, nbprocs, remaining_cpuset,
+                             HWLOC_GROUP_KIND_AMD_COMPUTE_UNIT, 0);
-			   TILE, "Tile",
+      }
-			   HWLOC_GROUP_KIND_INTEL_TILE, 0);
+      if (data->found_module_ids) {
+        /* Look for Intel Modules inside packages */
+        hwloc_bitmap_copy(remaining_cpuset, complete_cpuset);
+        hwloc_x86_add_groups(topology, infos, nbprocs, remaining_cpuset,
+                             MODULE, "Module",
+                             HWLOC_GROUP_KIND_INTEL_MODULE, 0);
+      }
+      if (data->found_tile_ids) {
+        /* Look for Intel Tiles inside packages */
+        hwloc_bitmap_copy(remaining_cpuset, complete_cpuset);
+        hwloc_x86_add_groups(topology, infos, nbprocs, remaining_cpuset,
+                             TILE, "Tile",
+                             HWLOC_GROUP_KIND_INTEL_TILE, 0);
+      }
 
       /* Look for unknown objects */
       if (infos[one].otherids) {
@@ -1060,7 +1160,8 @@ static void summarize(struct hwloc_backend *backend, struct procinfo *infos, uns
     }
   }
 
-  if (hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_DIE)) {
+  if (data->found_die_ids
+      && hwloc_filter_check_keep_object_type(topology, HWLOC_OBJ_DIE)) {
     /* Look for Intel Dies inside packages */
     if (fulldiscovery) {
       hwloc_bitmap_t die_cpuset;
@@ -1218,6 +1319,18 @@ static void summarize(struct hwloc_backend *backend, struct procinfo *infos, uns
 	    }
 	  }
 	  cache = hwloc_alloc_setup_object(topology, otype, HWLOC_UNKNOWN_INDEX);
+          /* We don't specify the os_index of caches because we want to be
+           * 100% sure they are identical to what the Linux kernel reports
+           * (so that things like resctrl work).
+           * However, vendor/model-specific quirks in the x86 code above
+           * make this difficult.
+           *
+           * Caveat: if the x86 backend is used on Linux to avoid kernel bugs,
+           * IDs won't be available to resctrl users. But resctrl heavily
+           * relies on the kernel x86 discovery being non-buggy anyway.
+           *
+           * TODO: make this optional? or only disable it on Linux?
+           */
 	  cache->attr->cache.depth = level;
 	  cache->attr->cache.size = infos[i].cache[l].size;
 	  cache->attr->cache.linesize = infos[i].cache[l].linesize;
@@ -1247,7 +1360,8 @@ static int
 look_procs(struct hwloc_backend *backend, struct procinfo *infos, unsigned long flags,
 	   unsigned highest_cpuid, unsigned highest_ext_cpuid, unsigned *features, enum cpuid_type cpuid_type,
 	   int (*get_cpubind)(hwloc_topology_t topology, hwloc_cpuset_t set, int flags),
-	   int (*set_cpubind)(hwloc_topology_t topology, hwloc_const_cpuset_t set, int flags))
+	   int (*set_cpubind)(hwloc_topology_t topology, hwloc_const_cpuset_t set, int flags),
+           hwloc_bitmap_t restrict_set)
 {
   struct hwloc_x86_backend_data_s *data = backend->private_data;
   struct hwloc_topology *topology = backend->topology;
@@ -1267,6 +1381,12 @@ look_procs(struct hwloc_backend *backend, struct procinfo *infos, unsigned long
 
   for (i = 0; i < nbprocs; i++) {
     struct cpuiddump *src_cpuiddump = NULL;
+
+    if (restrict_set && !hwloc_bitmap_isset(restrict_set, i)) {
+      /* skip this CPU outside of the binding mask */
+      continue;
+    }
+
     if (data->src_cpuiddump_path) {
       src_cpuiddump = cpuiddump_read(data->src_cpuiddump_path, i);
       if (!src_cpuiddump)
@@ -1296,40 +1416,45 @@ look_procs(struct hwloc_backend *backend, struct procinfo *infos, unsigned long
   if (data->apicid_unique) {
     summarize(backend, infos, flags);
 
-    if (has_hybrid(features)) {
+    if (data->is_hybrid
+        && !(topology->flags & HWLOC_TOPOLOGY_FLAG_NO_CPUKINDS)) {
       /* use hybrid info for cpukinds */
-      hwloc_bitmap_t atomset = hwloc_bitmap_alloc();
+      if (cpuid_type == intel) {
-      hwloc_bitmap_t coreset = hwloc_bitmap_alloc();
+        /* Hybrid Intel */
-      for(i=0; i<nbprocs; i++) {
+        hwloc_bitmap_t atomset = hwloc_bitmap_alloc();
-        if (infos[i].hybridcoretype == 0x20)
+        hwloc_bitmap_t coreset = hwloc_bitmap_alloc();
-          hwloc_bitmap_set(atomset, i);
+        for(i=0; i<nbprocs; i++) {
-        else if (infos[i].hybridcoretype == 0x40)
+          if (infos[i].hybridcoretype == 0x20)
-          hwloc_bitmap_set(coreset, i);
+            hwloc_bitmap_set(atomset, i);
-      }
+          else if (infos[i].hybridcoretype == 0x40)
-      /* register IntelAtom set if any */
+            hwloc_bitmap_set(coreset, i);
-      if (!hwloc_bitmap_iszero(atomset)) {
+        }
-        struct hwloc_info_s infoattr;
+        /* register IntelAtom set if any */
-        infoattr.name = (char *) "CoreType";
+        if (!hwloc_bitmap_iszero(atomset)) {
-        infoattr.value = (char *) "IntelAtom";
+          struct hwloc_info_s infoattr;
-        hwloc_internal_cpukinds_register(topology, atomset, HWLOC_CPUKIND_EFFICIENCY_UNKNOWN, &infoattr, 1, 0);
+          infoattr.name = (char *) "CoreType";
-        /* the cpuset is given to the callee */
+          infoattr.value = (char *) "IntelAtom";
-      } else {
+          hwloc_internal_cpukinds_register(topology, atomset, HWLOC_CPUKIND_EFFICIENCY_UNKNOWN, &infoattr, 1, 0);
-        hwloc_bitmap_free(atomset);
+          /* the cpuset is given to the callee */
-      }
+        } else {
-      /* register IntelCore set if any */
+          hwloc_bitmap_free(atomset);
-      if (!hwloc_bitmap_iszero(coreset)) {
+        }
-        struct hwloc_info_s infoattr;
+        /* register IntelCore set if any */
-        infoattr.name = (char *) "CoreType";
+        if (!hwloc_bitmap_iszero(coreset)) {
-        infoattr.value = (char *) "IntelCore";
+          struct hwloc_info_s infoattr;
-        hwloc_internal_cpukinds_register(topology, coreset, HWLOC_CPUKIND_EFFICIENCY_UNKNOWN, &infoattr, 1, 0);
+          infoattr.name = (char *) "CoreType";
-        /* the cpuset is given to the callee */
+          infoattr.value = (char *) "IntelCore";
-      } else {
+          hwloc_internal_cpukinds_register(topology, coreset, HWLOC_CPUKIND_EFFICIENCY_UNKNOWN, &infoattr, 1, 0);
-        hwloc_bitmap_free(coreset);
+          /* the cpuset is given to the callee */
+        } else {
+          hwloc_bitmap_free(coreset);
+        }
       }
     }
+  } else {
+    hwloc_debug("x86 APIC IDs aren't unique, x86 discovery ignored.\n");
+    /* do nothing and return success, so that the caller does nothing either */
   }
-  /* if !data->apicid_unique, do nothing and return success, so that the caller does nothing either */
-
   return 0;
 }
 
@@ -1400,12 +1525,21 @@ static
 int hwloc_look_x86(struct hwloc_backend *backend, unsigned long flags)
 {
   struct hwloc_x86_backend_data_s *data = backend->private_data;
+  struct hwloc_topology *topology = backend->topology;
   unsigned nbprocs = data->nbprocs;
   unsigned eax, ebx, ecx = 0, edx;
   unsigned i;
   unsigned highest_cpuid;
   unsigned highest_ext_cpuid;
-  /* This stores cpuid features with the same indexing as Linux */
+  /* This stores cpuid features with the same indexing as Linux:
+   * [0] = 0x1 edx
+   * [1] = 0x80000001 edx
+   * [4] = 0x1 ecx
+   * [6] = 0x80000001 ecx
+   * [9] = 0x7/0 ebx
+   * [16] = 0x7/0 ecx
+   * [18] = 0x7/0 edx
+   */
   unsigned features[19] = { 0 };
   struct procinfo *infos = NULL;
   enum cpuid_type cpuid_type = unknown;
@@ -1415,9 +1549,21 @@ int hwloc_look_x86(struct hwloc_backend *backend, unsigned long flags)
   struct hwloc_topology_membind_support memsupport __hwloc_attribute_unused;
   int (*get_cpubind)(hwloc_topology_t topology, hwloc_cpuset_t set, int flags) = NULL;
   int (*set_cpubind)(hwloc_topology_t topology, hwloc_const_cpuset_t set, int flags) = NULL;
+  hwloc_bitmap_t restrict_set = NULL;
   struct cpuiddump *src_cpuiddump = NULL;
   int ret = -1;
 
+  /* check if binding works */
+  memset(&hooks, 0, sizeof(hooks));
+  support.membind = &memsupport;
+  /* We could just copy the main hooks (except in some corner cases),
+   * but the current overhead is negligible, so just always reget them.
+   */
+  hwloc_set_native_binding_hooks(&hooks, &support);
+  /* in theory, those are only needed if !data->src_cpuiddump_path || HWLOC_TOPOLOGY_FLAG_RESTRICT_TO_BINDING
+   * but that's the vast majority of cases anyway, and the overhead is very small.
+   */
+
   if (data->src_cpuiddump_path) {
     /* Just read cpuid from the dump (implies !topology->is_thissystem by default) */
     src_cpuiddump = cpuiddump_read(data->src_cpuiddump_path, 0);
@@ -1430,13 +1576,6 @@ int hwloc_look_x86(struct hwloc_backend *backend, unsigned long flags)
      * we may still force use this backend when debugging with !thissystem.
      */
 
-    /* check if binding works */
-    memset(&hooks, 0, sizeof(hooks));
-    support.membind = &memsupport;
-    /* We could just copy the main hooks (except in some corner cases),
-     * but the current overhead is negligible, so just always reget them.
-     */
-    hwloc_set_native_binding_hooks(&hooks, &support);
     if (hooks.get_thisthread_cpubind && hooks.set_thisthread_cpubind) {
       get_cpubind = hooks.get_thisthread_cpubind;
       set_cpubind = hooks.set_thisthread_cpubind;
@@ -1456,6 +1595,20 @@ int hwloc_look_x86(struct hwloc_backend *backend, unsigned long flags)
     }
   }
 
+  if (topology->flags & HWLOC_TOPOLOGY_FLAG_RESTRICT_TO_CPUBINDING) {
+    restrict_set = hwloc_bitmap_alloc();
+    if (!restrict_set)
+      goto out;
+    if (hooks.get_thisproc_cpubind)
+      hooks.get_thisproc_cpubind(topology, restrict_set, 0);
+    else if (hooks.get_thisthread_cpubind)
+      hooks.get_thisthread_cpubind(topology, restrict_set, 0);
+    if (hwloc_bitmap_iszero(restrict_set)) {
+      hwloc_bitmap_free(restrict_set);
+      restrict_set = NULL;
+    }
+  }
+
   if (!src_cpuiddump && !hwloc_have_x86_cpuid())
     goto out;
 
@@ -1506,6 +1659,7 @@ int hwloc_look_x86(struct hwloc_backend *backend, unsigned long flags)
     ecx = 0;
     cpuid_or_from_dump(&eax, &ebx, &ecx, &edx, src_cpuiddump);
     features[9] = ebx;
+    features[16] = ecx;
     features[18] = edx;
   }
 
@@ -1520,7 +1674,7 @@ int hwloc_look_x86(struct hwloc_backend *backend, unsigned long flags)
 
   ret = look_procs(backend, infos, flags,
 		   highest_cpuid, highest_ext_cpuid, features, cpuid_type,
-		   get_cpubind, set_cpubind);
+		   get_cpubind, set_cpubind, restrict_set);
   if (!ret)
     /* success, we're done */
     goto out_with_os_state;
@@ -1545,6 +1699,7 @@ out_with_infos:
   }
 
 out:
+  hwloc_bitmap_free(restrict_set);
   if (src_cpuiddump)
     cpuiddump_free(src_cpuiddump);
   return ret;
@@ -1561,6 +1716,11 @@ hwloc_x86_discover(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
 
   assert(dstatus->phase == HWLOC_DISC_PHASE_CPU);
 
+  if (topology->flags & HWLOC_TOPOLOGY_FLAG_DONT_CHANGE_BINDING) {
+    /* TODO: Things would work if there's a single PU, no need to rebind */
+    return 0;
+  }
+
   if (getenv("HWLOC_X86_TOPOEXT_NUMANODES")) {
     flags |= HWLOC_X86_DISC_FLAG_TOPOEXT_NUMANODES;
   }
@@ -1587,7 +1747,8 @@ hwloc_x86_discover(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
   }
 
   if (topology->levels[0][0]->cpuset) {
-    /* somebody else discovered things */
+    /* somebody else discovered things, reconnect levels so that we can look at them */
+    hwloc__reconnect(topology, 0);
     if (topology->nb_levels == 2 && topology->level_nbobjects[1] == data->nbprocs) {
       /* only PUs were discovered, as much as we would, complete the topology with everything else */
       alreadypus = 1;
@@ -1595,7 +1756,6 @@ hwloc_x86_discover(struct hwloc_backend *backend, struct hwloc_disc_status *dsta
     }
 
     /* several object types were added, we can't easily complete, just do partial discovery */
-    hwloc_topology_reconnect(topology, 0);
     ret = hwloc_look_x86(backend, flags);
     if (ret)
       hwloc_obj_add_info(topology->levels[0][0], "Backend", "x86");
@@ -1651,17 +1811,17 @@ hwloc_x86_check_cpuiddump_input(const char *src_cpuiddump_path, hwloc_bitmap_t s
   sprintf(path, "%s/hwloc-cpuid-info", src_cpuiddump_path);
   file = fopen(path, "r");
   if (!file) {
-    fprintf(stderr, "Couldn't open dumped cpuid summary %s\n", path);
+    fprintf(stderr, "hwloc/x86: Couldn't open dumped cpuid summary %s\n", path);
     goto out_with_path;
   }
   if (!fgets(line, sizeof(line), file)) {
-    fprintf(stderr, "Found read dumped cpuid summary in %s\n", path);
+    fprintf(stderr, "hwloc/x86: Found read dumped cpuid summary in %s\n", path);
     fclose(file);
     goto out_with_path;
   }
   fclose(file);
-  if (strcmp(line, "Architecture: x86\n")) {
+  if (strncmp(line, "Architecture: x86", 17)) {
-    fprintf(stderr, "Found non-x86 dumped cpuid summary in %s: %s\n", path, line);
+    fprintf(stderr, "hwloc/x86: Found non-x86 dumped cpuid summary in %s: %s\n", path, line);
     goto out_with_path;
   }
   free(path);
@@ -1673,19 +1833,19 @@ hwloc_x86_check_cpuiddump_input(const char *src_cpuiddump_path, hwloc_bitmap_t s
       if (!*end)
 	hwloc_bitmap_set(set, idx);
       else
-	fprintf(stderr, "Ignoring invalid dirent `%s' in dumped cpuid directory `%s'\n",
+	fprintf(stderr, "hwloc/x86: Ignoring invalid dirent `%s' in dumped cpuid directory `%s'\n",
 		dirent->d_name, src_cpuiddump_path);
     }
   }
   closedir(dir);
 
   if (hwloc_bitmap_iszero(set)) {
-    fprintf(stderr, "Did not find any valid pu%%u entry in dumped cpuid directory `%s'\n",
+    fprintf(stderr, "hwloc/x86: Did not find any valid pu%%u entry in dumped cpuid directory `%s'\n",
 	    src_cpuiddump_path);
     return -1;
   } else if (hwloc_bitmap_last(set) != hwloc_bitmap_weight(set) - 1) {
     /* The x86 backends enforces contigous set of PUs starting at 0 so far */
-    fprintf(stderr, "Found non-contigous pu%%u range in dumped cpuid directory `%s'\n",
+    fprintf(stderr, "hwloc/x86: Found non-contigous pu%%u range in dumped cpuid directory `%s'\n",
 	    src_cpuiddump_path);
     return -1;
   }
@@ -1737,9 +1897,15 @@ hwloc_x86_component_instantiate(struct hwloc_topology *topology,
 
   /* default values */
   data->is_knl = 0;
+  data->is_hybrid = 0;
   data->apicid_set = hwloc_bitmap_alloc();
   data->apicid_unique = 1;
   data->src_cpuiddump_path = NULL;
+  data->found_die_ids = 0;
+  data->found_complex_ids = 0;
+  data->found_unit_ids = 0;
+  data->found_module_ids = 0;
+  data->found_tile_ids = 0;
 
   src_cpuiddump_path = getenv("HWLOC_CPUID_PATH");
   if (src_cpuiddump_path) {
@@ -1750,7 +1916,7 @@ hwloc_x86_component_instantiate(struct hwloc_topology *topology,
       assert(!hwloc_bitmap_iszero(set)); /* enforced by hwloc_x86_check_cpuiddump_input() */
       data->nbprocs = hwloc_bitmap_weight(set);
     } else {
-      fprintf(stderr, "Ignoring dumped cpuid directory.\n");
+      fprintf(stderr, "hwloc/x86: Ignoring dumped cpuid directory.\n");
     }
     hwloc_bitmap_free(set);
   }

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2024 Inria.  All rights reserved.
  * Copyright © 2009-2011 Université Bordeaux
  * Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -41,7 +41,7 @@ typedef struct hwloc__nolibxml_import_state_data_s {
 static char *
 hwloc__nolibxml_import_ignore_spaces(char *buffer)
 {
-  return buffer + strspn(buffer, " \t\n");
+  return buffer + strspn(buffer, " \t\n\r");
 }
 
 static int
@@ -411,12 +411,12 @@ hwloc_nolibxml_backend_init(struct hwloc_xml_backend_data_s *bdata,
   bdata->data = nbdata;
 
   if (xmlbuffer) {
-    nbdata->buffer = malloc(xmlbuflen+1);
+    nbdata->buffer = malloc(xmlbuflen);
     if (!nbdata->buffer)
       goto out_with_nbdata;
-    nbdata->buflen = xmlbuflen+1;
+    nbdata->buflen = xmlbuflen;
     memcpy(nbdata->buffer, xmlbuffer, xmlbuflen);
-    nbdata->buffer[xmlbuflen] = '\0';
+    nbdata->buffer[xmlbuflen-1] = '\0'; /* make sure it's there as requested in the API */
 
   } else {
     int err = hwloc_nolibxml_read_file(xmlpath, &nbdata->buffer, &nbdata->buflen);
@@ -453,8 +453,9 @@ hwloc_nolibxml_import_diff(struct hwloc__xml_import_state_s *state,
     buffer = malloc(xmlbuflen);
     if (!buffer)
       goto out;
-    memcpy(buffer, xmlbuffer, xmlbuflen);
     buflen = xmlbuflen;
+    memcpy(buffer, xmlbuffer, xmlbuflen);
+    buffer[xmlbuflen-1] = '\0'; /* make sure it's there as requested in the API */
 
   } else {
     ret = hwloc_nolibxml_read_file(xmlpath, &buffer, &buflen);

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

@@ -1,6 +1,6 @@
 /*
  * Copyright © 2009 CNRS
- * Copyright © 2009-2020 Inria.  All rights reserved.
+ * Copyright © 2009-2025 Inria.  All rights reserved.
  * Copyright © 2009-2011, 2020 Université Bordeaux
  * Copyright © 2009-2018 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -123,6 +123,17 @@ hwloc__xml_import_object_attr(struct hwloc_topology *topology,
       fprintf(stderr, "%s: unexpected zero gp_index, topology may be invalid\n", state->global->msgprefix);
     if (obj->gp_index >= topology->next_gp_index)
       topology->next_gp_index = obj->gp_index + 1;
+  } else if (!strcmp(name, "id")) { /* forward compat */
+    if (!strncmp(value, "obj", 3)) {
+      obj->gp_index = strtoull(value+3, NULL, 10);
+      if (!obj->gp_index && hwloc__xml_verbose())
+        fprintf(stderr, "%s: unexpected zero id, topology may be invalid\n", state->global->msgprefix);
+      if (obj->gp_index >= topology->next_gp_index)
+        topology->next_gp_index = obj->gp_index + 1;
+    } else {
+      if (hwloc__xml_verbose())
+        fprintf(stderr, "%s: unexpected id `%s' not-starting with `obj', ignoring\n", state->global->msgprefix, value);
+    }
   } else if (!strcmp(name, "cpuset")) {
     if (!obj->cpuset)
       obj->cpuset = hwloc_bitmap_alloc();
@@ -192,8 +203,9 @@ hwloc__xml_import_object_attr(struct hwloc_topology *topology,
 	  || lvalue == HWLOC_OBJ_CACHE_INSTRUCTION)
 	obj->attr->cache.type = (hwloc_obj_cache_type_t) lvalue;
       else
-	fprintf(stderr, "%s: ignoring invalid cache_type attribute %lu\n",
+        if (hwloc__xml_verbose())
-		state->global->msgprefix, lvalue);
+          fprintf(stderr, "%s: ignoring invalid cache_type attribute %lu\n",
+                  state->global->msgprefix, lvalue);
     } else if (hwloc__xml_verbose())
       fprintf(stderr, "%s: ignoring cache_type attribute for non-cache object type\n",
 	      state->global->msgprefix);
@@ -242,7 +254,7 @@ hwloc__xml_import_object_attr(struct hwloc_topology *topology,
   else if (!strcmp(name, "dont_merge")) {
     unsigned long lvalue = strtoul(value, NULL, 10);
     if (obj->type == HWLOC_OBJ_GROUP)
-      obj->attr->group.dont_merge = lvalue;
+      obj->attr->group.dont_merge = (unsigned char) lvalue;
     else if (hwloc__xml_verbose())
       fprintf(stderr, "%s: ignoring dont_merge attribute for non-group object type\n",
 	      state->global->msgprefix);
@@ -262,8 +274,8 @@ hwloc__xml_import_object_attr(struct hwloc_topology *topology,
 #ifndef HWLOC_HAVE_32BITS_PCI_DOMAIN
       } else if (domain > 0xffff) {
 	static int warned = 0;
-	if (!warned && !hwloc_hide_errors())
+	if (!warned && HWLOC_SHOW_ALL_ERRORS())
-	  fprintf(stderr, "Ignoring PCI device with non-16bit domain.\nPass --enable-32bits-pci-domain to configure to support such devices\n(warning: it would break the library ABI, don't enable unless really needed).\n");
+	  fprintf(stderr, "hwloc/xml: Ignoring PCI device with non-16bit domain.\nPass --enable-32bits-pci-domain to configure to support such devices\n(warning: it would break the library ABI, don't enable unless really needed).\n");
 	warned = 1;
 	*ignore = 1;
 #endif
@@ -337,6 +349,7 @@ hwloc__xml_import_object_attr(struct hwloc_topology *topology,
       } else {
 	obj->attr->bridge.upstream_type = (hwloc_obj_bridge_type_t) upstream_type;
 	obj->attr->bridge.downstream_type = (hwloc_obj_bridge_type_t) downstream_type;
+        /* FIXME verify that upstream/downstream type is valid */
       };
       break;
     }
@@ -361,12 +374,13 @@ hwloc__xml_import_object_attr(struct hwloc_topology *topology,
 #ifndef HWLOC_HAVE_32BITS_PCI_DOMAIN
       } else if (domain > 0xffff) {
 	static int warned = 0;
-	if (!warned && !hwloc_hide_errors())
+	if (!warned && HWLOC_SHOW_ALL_ERRORS())
-	  fprintf(stderr, "Ignoring bridge to PCI with non-16bit domain.\nPass --enable-32bits-pci-domain to configure to support such devices\n(warning: it would break the library ABI, don't enable unless really needed).\n");
+	  fprintf(stderr, "hwloc/xml: Ignoring bridge to PCI with non-16bit domain.\nPass --enable-32bits-pci-domain to configure to support such devices\n(warning: it would break the library ABI, don't enable unless really needed).\n");
 	warned = 1;
 	*ignore = 1;
 #endif
       } else {
+        /* FIXME verify that downstream type vs pci info are valid */
 	obj->attr->bridge.downstream.pci.domain = domain;
 	obj->attr->bridge.downstream.pci.secondary_bus = secbus;
 	obj->attr->bridge.downstream.pci.subordinate_bus = subbus;
@@ -401,6 +415,20 @@ hwloc__xml_import_object_attr(struct hwloc_topology *topology,
     }
   }
 
+  else if (!strcmp(name, "numanode_type")) {
+    switch (obj->type) {
+    case HWLOC_OBJ_NUMANODE: {
+      /* ignored for now, here for possible forward compat */
+      break;
+    }
+    default:
+      if (hwloc__xml_verbose())
+	fprintf(stderr, "%s: ignoring numanode_type attribute for non-NUMA object\n",
+		state->global->msgprefix);
+      break;
+    }
+  }
+
   else if (data->version_major < 2) {
     /************************
      * deprecated from 1.x
@@ -548,7 +576,13 @@ hwloc__xml_import_pagetype(hwloc_topology_t topology __hwloc_attribute_unused, s
     char *attrname, *attrvalue;
     if (state->global->next_attr(state, &attrname, &attrvalue) < 0)
       break;
-    if (!strcmp(attrname, "size"))
+    if (!strcmp(attrname, "info")) {
+      char *infoname, *infovalue;
+      int ret = hwloc___xml_import_info(&infoname, &infovalue, state);
+      if (ret < 0)
+        return -1;
+      /* ignored */
+    } else if (!strcmp(attrname, "size"))
       size = strtoull(attrvalue, NULL, 10);
     else if (!strcmp(attrname, "count"))
       count = strtoull(attrvalue, NULL, 10);
@@ -852,14 +886,23 @@ hwloc__xml_import_object(hwloc_topology_t topology,
 	  /* deal with possible future type */
 	  obj->type = HWLOC_OBJ_GROUP;
 	  obj->attr->group.kind = HWLOC_GROUP_KIND_INTEL_MODULE;
-	} else if (!strcasecmp(attrvalue, "MemCache")) {
+	} else if (!strcasecmp(attrvalue, "Cluster")) {
+	  /* deal with possible future type */
+	  obj->type = HWLOC_OBJ_GROUP;
+	  obj->attr->group.kind = HWLOC_GROUP_KIND_LINUX_CLUSTER;
+	}
+#if 0
+        /* reenable if there's ever a future type that should be ignored without being an error */
+        else if (!strcasecmp(attrvalue, "MemCache")) {
 	  /* ignore possible future type */
 	  obj->type = _HWLOC_OBJ_FUTURE;
 	  ignored = 1;
 	  if (hwloc__xml_verbose())
 	    fprintf(stderr, "%s: %s object not-supported, will be ignored\n",
 		    state->global->msgprefix, attrvalue);
-	} else {
+	}
+#endif
+        else {
 	  if (hwloc__xml_verbose())
 	    fprintf(stderr, "%s: unrecognized object type string %s\n",
 		    state->global->msgprefix, attrvalue);
@@ -934,22 +977,22 @@ hwloc__xml_import_object(hwloc_topology_t topology,
     if (hwloc__obj_type_is_normal(obj->type)) {
       if (!hwloc__obj_type_is_normal(parent->type)) {
 	if (hwloc__xml_verbose())
-	  fprintf(stderr, "normal object %s cannot be child of non-normal parent %s\n",
+	  fprintf(stderr, "%s: normal object %s cannot be child of non-normal parent %s\n",
-		  hwloc_obj_type_string(obj->type), hwloc_obj_type_string(parent->type));
+		  state->global->msgprefix, hwloc_obj_type_string(obj->type), hwloc_obj_type_string(parent->type));
 	goto error_with_object;
       }
     } else if (hwloc__obj_type_is_memory(obj->type)) {
       if (hwloc__obj_type_is_io(parent->type) || HWLOC_OBJ_MISC == parent->type) {
 	if (hwloc__xml_verbose())
-	  fprintf(stderr, "Memory object %s cannot be child of non-normal-or-memory parent %s\n",
+	  fprintf(stderr, "%s: Memory object %s cannot be child of non-normal-or-memory parent %s\n",
-		  hwloc_obj_type_string(obj->type), hwloc_obj_type_string(parent->type));
+		  state->global->msgprefix, hwloc_obj_type_string(obj->type), hwloc_obj_type_string(parent->type));
 	goto error_with_object;
       }
     } else if (hwloc__obj_type_is_io(obj->type)) {
       if (hwloc__obj_type_is_memory(parent->type) || HWLOC_OBJ_MISC == parent->type) {
 	if (hwloc__xml_verbose())
-	  fprintf(stderr, "I/O object %s cannot be child of non-normal-or-I/O parent %s\n",
+	  fprintf(stderr, "%s: I/O object %s cannot be child of non-normal-or-I/O parent %s\n",
-		  hwloc_obj_type_string(obj->type), hwloc_obj_type_string(parent->type));
+		  state->global->msgprefix, hwloc_obj_type_string(obj->type), hwloc_obj_type_string(parent->type));
 	goto error_with_object;
       }
     }
@@ -1146,6 +1189,48 @@ hwloc__xml_import_object(hwloc_topology_t topology,
     data->last_numanode = obj;
   }
 
+  /* 3.0 forward compatibility */
+  if (data->version_major >= 3 && obj->type == HWLOC_OBJ_OS_DEVICE) {
+    /* osdev.type changed into bitmak in 3.0 */
+    if (obj->attr->osdev.type & 3 /* STORAGE|MEMORY for BLOCK */) {
+      obj->attr->osdev.type = HWLOC_OBJ_OSDEV_BLOCK;
+    } else if (obj->attr->osdev.type & 8 /* COPROC for COPROC and rsmi/nvml GPUs */) {
+      if (obj->subtype && (!strcmp(obj->subtype, "RSMI") || !strcmp(obj->subtype, "NVML")))
+        obj->attr->osdev.type = HWLOC_OBJ_OSDEV_GPU;
+      else
+        obj->attr->osdev.type = HWLOC_OBJ_OSDEV_COPROC;
+    } else if (obj->attr->osdev.type & 4 /* GPU for non-COPROC GPUs */) {
+      obj->attr->osdev.type = HWLOC_OBJ_OSDEV_GPU;
+    } else if (obj->attr->osdev.type & 32 /* OFED */) {
+      obj->attr->osdev.type = HWLOC_OBJ_OSDEV_OPENFABRICS;
+    } else if (obj->attr->osdev.type & 16 /* NET for NET and BXI v2-fake-OFED */) {
+      if (obj->subtype && !strcmp(obj->subtype, "BXI"))
+        obj->attr->osdev.type = HWLOC_OBJ_OSDEV_OPENFABRICS;
+      else
+        obj->attr->osdev.type = HWLOC_OBJ_OSDEV_NETWORK;
+    } else if (obj->attr->osdev.type & 64 /* DMA */) {
+      obj->attr->osdev.type = HWLOC_OBJ_OSDEV_DMA;
+    } else { /* none or unknown */
+      obj->attr->osdev.type = (hwloc_obj_osdev_type_t)  -1;
+    }
+    /* Backend info only in root */
+    if (obj->subtype && !hwloc_obj_get_info_by_name(obj, "Backend")) {
+      if (!strcmp(obj->subtype, "CUDA")) {
+        hwloc_obj_add_info(obj, "Backend", "CUDA");
+      } else if  (!strcmp(obj->subtype, "NVML")) {
+        hwloc_obj_add_info(obj, "Backend", "NVML");
+      } else if  (!strcmp(obj->subtype, "OpenCL")) {
+        hwloc_obj_add_info(obj, "Backend", "OpenCL");
+      } else if  (!strcmp(obj->subtype, "RSMI")) {
+        hwloc_obj_add_info(obj, "Backend", "RSMI");
+      } else if  (!strcmp(obj->subtype, "LevelZero")) {
+        hwloc_obj_add_info(obj, "Backend", "LevelZero");
+      } else if  (!strcmp(obj->subtype, "Display")) {
+        hwloc_obj_add_info(obj, "Backend", "GL");
+      }
+    }
+  }
+
   if (!hwloc_filter_check_keep_object(topology, obj)) {
     /* Ignore this object instead of inserting it.
      *
@@ -1232,7 +1317,7 @@ hwloc__xml_import_object(hwloc_topology_t topology,
 	/* next should be before cur */
 	if (!childrengotignored) {
 	  static int reported = 0;
-	  if (!reported && !hwloc_hide_errors()) {
+	  if (!reported && HWLOC_SHOW_CRITICAL_ERRORS()) {
 	    hwloc__xml_import_report_outoforder(topology, next, cur);
 	    reported = 1;
 	  }
@@ -1282,7 +1367,7 @@ hwloc__xml_v2import_support(hwloc_topology_t topology,
     HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_support) == 4*sizeof(void*));
     HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_discovery_support) == 6);
     HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_cpubind_support) == 11);
-    HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_membind_support) == 15);
+    HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_membind_support) == 16);
     HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_misc_support) == 1);
 #endif
 
@@ -1316,6 +1401,7 @@ hwloc__xml_v2import_support(hwloc_topology_t topology,
     else DO(membind,firsttouch_membind);
     else DO(membind,bind_membind);
     else DO(membind,interleave_membind);
+    else DO(membind,weighted_interleave_membind);
     else DO(membind,nexttouch_membind);
     else DO(membind,migrate_membind);
     else DO(membind,get_area_memlocation);
@@ -1374,6 +1460,10 @@ hwloc__xml_v2import_distances(hwloc_topology_t topology,
     }
     else if (!strcmp(attrname, "kind")) {
       kind = strtoul(attrvalue, NULL, 10);
+      /* forward compat with "HOPS" kind in v3 */
+      if (kind & (1UL<<5))
+        /* hops becomes latency */
+        kind = (kind & ~(1UL<<5)) | HWLOC_DISTANCES_KIND_MEANS_LATENCY;
     }
     else if (!strcmp(attrname, "name")) {
       name = attrvalue;
@@ -1419,7 +1509,14 @@ hwloc__xml_v2import_distances(hwloc_topology_t topology,
     if (ret <= 0)
       break;
 
-    if (!strcmp(tag, "indexes"))
+    if (!strcmp(tag, "info")) {
+      char *infoname, *infovalue;
+      ret = hwloc___xml_import_info(&infoname, &infovalue, state);
+      if (ret < 0)
+        goto out_with_arrays;
+      /* ignored */
+      continue;
+    } else if (!strcmp(tag, "indexes"))
       is_index = 1;
     else if (!strcmp(tag, "u64values"))
       is_u64values = 1;
@@ -1462,6 +1559,9 @@ hwloc__xml_v2import_distances(hwloc_topology_t topology,
 	unsigned long long u;
 	if (heterotypes) {
 	  hwloc_obj_type_t t = HWLOC_OBJ_TYPE_NONE;
+          if (!*tmp)
+            /* reached the end of this indexes attribute */
+            break;
 	  if (hwloc_type_sscanf(tmp, &t, NULL, 0) < 0) {
 	    if (hwloc__xml_verbose())
 	      fprintf(stderr, "%s: %s with unrecognized heterogeneous type %s\n",
@@ -1562,7 +1662,10 @@ hwloc__xml_v2import_distances(hwloc_topology_t topology,
     }
   }
 
-  hwloc_internal_distances_add_by_index(topology, name, unique_type, different_types, nbobjs, indexes, u64values, kind, 0);
+  if (topology->flags & HWLOC_TOPOLOGY_FLAG_NO_DISTANCES)
+    goto out_ignore;
+
+  hwloc_internal_distances_add_by_index(topology, name, unique_type, different_types, nbobjs, indexes, u64values, kind, 0 /* assume grouping was applied when this matrix was discovered before exporting to XML */);
 
   /* prevent freeing below */
   indexes = NULL;
@@ -1716,7 +1819,8 @@ hwloc__xml_import_memattr(hwloc_topology_t topology,
     }
   }
 
-  if (name && flags != (unsigned long) -1) {
+  if (name && flags != (unsigned long) -1
+      && !(topology->flags & HWLOC_TOPOLOGY_FLAG_NO_MEMATTRS)) {
     hwloc_memattr_id_t _id;
 
     ret = hwloc_memattr_get_by_name(topology, name, &_id);
@@ -1745,6 +1849,10 @@ hwloc__xml_import_memattr(hwloc_topology_t topology,
 
     if (!strcmp(tag, "memattr_value")) {
       ret = hwloc__xml_import_memattr_value(topology, id, flags, &childstate);
+    } else if (!strcmp(tag, "info")) {
+      char *infoname, *infovalue;
+      ret = hwloc___xml_import_info(&infoname, &infovalue, &childstate);
+      /* ignored */
     } else {
       if (hwloc__xml_verbose())
         fprintf(stderr, "%s: memattr with unrecognized child %s\n",
@@ -1827,7 +1935,13 @@ hwloc__xml_import_cpukind(hwloc_topology_t topology,
     goto error;
   }
 
-  hwloc_internal_cpukinds_register(topology, cpuset, forced_efficiency, infos, nr_infos, HWLOC_CPUKINDS_REGISTER_FLAG_OVERWRITE_FORCED_EFFICIENCY);
+  if (topology->flags & HWLOC_TOPOLOGY_FLAG_NO_CPUKINDS) {
+    hwloc__free_infos(infos, nr_infos);
+    hwloc_bitmap_free(cpuset);
+  } else {
+    hwloc_internal_cpukinds_register(topology, cpuset, forced_efficiency, infos, nr_infos, HWLOC_CPUKINDS_REGISTER_FLAG_OVERWRITE_FORCED_EFFICIENCY);
+    hwloc__free_infos(infos, nr_infos);
+  }
 
   return state->global->close_tag(state);
 
@@ -2067,9 +2181,10 @@ hwloc_look_xml(struct hwloc_backend *backend, struct hwloc_disc_status *dstatus)
   if (ret < 0)
     goto failed;
 
-  if (data->version_major > 2) {
+  if (data->version_major > 3
+      || (data->version_major == 3 && data->version_minor > 0)) {
     if (hwloc__xml_verbose())
-      fprintf(stderr, "%s: cannot import XML version %u.%u > 2\n",
+      fprintf(stderr, "%s: cannot import XML version %u.%u > 3.0\n",
 	      data->msgprefix, data->version_major, data->version_minor);
     goto err;
   }
@@ -2117,6 +2232,13 @@ hwloc_look_xml(struct hwloc_backend *backend, struct hwloc_disc_status *dstatus)
         ret = hwloc__xml_import_cpukind(topology, &childstate);
         if (ret < 0)
           goto failed;
+      } else if (!strcmp(tag, "info")) {
+        char *infoname, *infovalue;
+        ret = hwloc___xml_import_info(&infoname, &infovalue, &childstate);
+        if (ret < 0)
+          goto failed;
+        /* move 3.x topology info back to the root object */
+        hwloc_obj_add_info(topology->levels[0][0], infoname, infovalue);
       } else {
 	if (hwloc__xml_verbose())
 	  fprintf(stderr, "%s: ignoring unknown tag `%s' after root object.\n",
@@ -2162,7 +2284,8 @@ done:
        * but it would require to have those objects in the original XML order (like the first_numanode cousin-list).
        * because the topology order can be different if some parents are ignored during load.
        */
-      if (nbobjs == data->nbnumanodes) {
+      if (nbobjs == data->nbnumanodes
+          && !(topology->flags & HWLOC_TOPOLOGY_FLAG_NO_DISTANCES)) {
 	hwloc_obj_t *objs = malloc(nbobjs*sizeof(hwloc_obj_t));
 	uint64_t *values = malloc(nbobjs*nbobjs*sizeof(*values));
         assert(data->nbnumanodes > 0); /* v1dist->nbobjs is >0 after import */
@@ -2644,7 +2767,8 @@ hwloc__xml_export_object_contents (hwloc__xml_export_state_t state, hwloc_topolo
 
       logical_to_v2array = malloc(nbobjs * sizeof(*logical_to_v2array));
       if (!logical_to_v2array) {
-	fprintf(stderr, "xml/export/v1: failed to allocated logical_to_v2array\n");
+        if (HWLOC_SHOW_ALL_ERRORS())
+          fprintf(stderr, "hwloc/xml/export/v1: failed to allocated logical_to_v2array\n");
 	continue;
       }
 
@@ -2818,6 +2942,7 @@ hwloc__xml_v1export_object_with_memory(hwloc__xml_export_state_t parentstate, hw
     /* child has sibling, we must add a Group around those memory children */
     hwloc_obj_t group = parentstate->global->v1_memory_group;
     parentstate->new_child(parentstate, &gstate, "object");
+    group->parent = obj->parent;
     group->cpuset = obj->cpuset;
     group->complete_cpuset = obj->complete_cpuset;
     group->nodeset = obj->nodeset;
@@ -2990,7 +3115,7 @@ hwloc__xml_v2export_support(hwloc__xml_export_state_t parentstate, hwloc_topolog
   HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_support) == 4*sizeof(void*));
   HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_discovery_support) == 6);
   HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_cpubind_support) == 11);
-  HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_membind_support) == 15);
+  HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_membind_support) == 16);
   HWLOC_BUILD_ASSERT(sizeof(struct hwloc_topology_misc_support) == 1);
 #endif
 
@@ -3035,6 +3160,7 @@ hwloc__xml_v2export_support(hwloc__xml_export_state_t parentstate, hwloc_topolog
   DO(membind,firsttouch_membind);
   DO(membind,bind_membind);
   DO(membind,interleave_membind);
+  DO(membind,weighted_interleave_membind);
   DO(membind,nexttouch_membind);
   DO(membind,migrate_membind);
   DO(membind,get_area_memlocation);
@@ -3116,9 +3242,11 @@ hwloc__xml_export_memattrs(hwloc__xml_export_state_t state, hwloc_topology_t top
       continue;
 
     imattr = &topology->memattrs[id];
-    if ((id == HWLOC_MEMATTR_ID_LATENCY || id == HWLOC_MEMATTR_ID_BANDWIDTH)
+    if (id < HWLOC_MEMATTR_ID_MAX && !imattr->nr_targets)
-        && !imattr->nr_targets)
+      /* no need to export standard attributes without any target,
-      /* no need to export target-less attributes for initial attributes, no release support attributes without those definitions */
+       * their definition is now standardized,
+       * the old hwloc importing this XML may recreate these attributes just like it would for a non-imported topology.
+       */
       continue;
 
     state->new_child(state, &mstate, "memattr");