News
Announcing PAPI 5.5.0
2016-09-14

PAPI 5.5 is now available. This release includes a new component that provides read and write access to the information and controls exposed via the Linux powercap interface.The PAPI powercap component supports measuring and capping power usage on recent Intel architectures.

We have added core support for Knights Landing (uncore support will be released later) as well as power monitoring via the RAPL and powercap components. 

As always, a very large Thank You goes out to all our collaborators and contributors!  This project would not be successful without your help!

The PAPI release can be downloaded from http://icl.cs.utk.edu/papi/software.

New Platforms:

  • Added  Knights Landing (KNL) core events and preset events.
  • Added Intel Broadwell/Skylake/Knights Landing RAPL support
  • Updated PAPI preset event support for Intel Broadwell/Skylake

New Component:

  • Powercap component:  PAPI now supports the Linux Power Capping Framework which exposes power capping devices and power measurement to user space via a sysfs virtual file system interface.  

Enhancements:

  • Add support for multiple flavors of POWER8 processors.
  • Force all processors to check event schedulability by checking that PAPI can successfully read the counters.
  • Support for Intel Broadwell-EP, Skylake, Goldmont, Haswell-EP inherited from libpfm4.
  • Shared memory object (.so) naming is made more limited so that minor updates do not break ABI compatibility.

Bug Fixes:

  • Improve testlib error messages if a component fails to initialize.
  • Fix _papi_hwi_postfix_calc parsing and robustness.
  • Clean build rules for CUDA sampling subcomponent.
  • Correct IBM Power7 and Power8 computation of PAPI_L1_DCA.
  • Eliminate the sole use of ftests_skip subroutine.
  • Correct the event string names for tenth.c.  
  • Have Fortran test support code report errors more clearly.
  • Cleanup output from libmsr component.
  • PAPI internal functions were marked as static to avoid exposing them externally.
  • Multiple component were fixed to make internal functions static where possible, to avoid exposing the functions as externally accessible entry points.
  • CUDA component configuration bug fixed.

 


Jun 24 2017 Admin Login