Open MPI logo

Hardware Locality announce Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Hardware Locality Announcements mailing list

Subject: [hwloc-announce] Hardware locality (hwloc) v1.9rc1 released
From: Brice Goglin (Brice.Goglin_at_[hidden])
Date: 2014-03-20 18:18:51


The Hardware Locality (hwloc) team is pleased to announce the first
release candidate for v1.9:

   http://www.open-mpi.org/projects/hwloc/

v1.9rc1 is the first milestone of a major feature release.
It adds a couple enhanced API functions, many object attributes for
better representing CPU and I/O device characteristics, and more.

* API
  + Add hwloc_obj_type_sscanf() to extend hwloc_obj_type_of_string() with
    type-specific attributes such as Cache/Group depth and Cache type.
    hwloc_obj_type_of_string() is moved to hwloc/deprecated.h.
  + Add hwloc_linux_get_tid_last_cpu_location() for retrieving the
    last CPU where a Linux thread given by TID ran.
  + Add hwloc_distrib() to extend the old hwloc_distribute[v]() functions.
    hwloc_distribute[v]() is moved to hwloc/deprecated.h.
* Backends
  + Add CPUVendor, CPUModelNumber and CPUFamilyNumber info attributes for
    x86, ia64 and Xeon Phi sockets on Linux, to extend the x86-specific
    support added in v1.8.1. Requested by Ralph Castain.
  + Add many CPU- and Platform-related info attributes on ARM and POWER
    platforms, in the Machine and Socket objects.
  + Add CUDA info attributes describing the number of multiprocessors and
    cores and the size of the global, shared and L2 cache memories in CUDA
    OS devices.
  + Add OpenCL info attributes describing the number of compute units and
    the global memory size in OpenCL OS devices.
  + The synthetic backend now accepts extended types such as L2Cache, L1i or
    Group3. lstopo also exports synthetic strings using these extended types.
* Tools
  + lstopo does not overwrite output files by default anymore.
    Pass -f or --force to enforce it.
  + lstopo now displays OpenCL, CUDA and Xeon Phi numbers of cores and
    memory sizes in the graphical output.
  + Add -e or --get-last-cpu-location to hwloc-ps to report where
    processes/threads run instead of where they are bound.
  + hwloc-ps reports locations as likely-more-useful objects such as Cores
    or Sockets instead of Caches when possible.
  + Add -e as a synonym to --get-last-cpu-location in hwloc-bind.
  + Add --reverse to hwloc-distrib to distribute using last objects first
    and singlify into last bits first. Thanks to Jirka Hladky for the
    suggestion.
  + hwloc-info now reports unified caches when looking for data or instruction
    cache ancestor objects.
* Misc
  + Add experimental Visual Studio support under contrib/windows.
    Thanks to Eloi Gaudry for his help and for providing the first draft.
  + Fix some overzealous assertions and warnings about the ordering of
    objects on a level with respect to cpusets. The ordering is only
    guaranteed for complete cpusets (based on the first bit in sets).

v1.9rc1 also contains some bug fixes that are waiting for a v1.8.2 stable
release that may or may not ever happen:

* Don't mix total and local memory when displaying verbose object attributes
  with hwloc_obj_attr_snprintf() or in lstopo.
* Fix a memory leak in the diff interface when adding a "too complex" entry.
* Fix lstopo export to stdout when specifying a Cairo-based output type
  with --of.
* Fix some memory leaks when importing xml diffs.
* Fix hwloc-bind failure on Windows when not using --pid.

--
Brice