PAPI 5.7 is now available. This release includes a new component, called "pcp", which interfaces to the Performance Co-Pilot (PCP). It enables PAPI users to monitor IBM POWER9 hardware performance events, particularly shared “NEST” events without root access.
This release also upgrades the (to date read-only) PAPI “nvml” component with write access to the information and controls exposed via the NVIDIA Management Library. The PAPI “nvml” component now supports both---measuring and capping power usage---on recent NVIDIA GPU architectures (e.g. V100).
We have added power monitoring as well as PMU support for recent Intel architectures such as Cascade Lake, Kaby Lake, Skylake, and Knights Mill (KNM). Furthermore, measuring power usage for AMD Fam17h chips is now available via the “rapl” component.
The PAPI release can be downloaded from http://icl.utk.edu/papi/software.
For specific and detailed information on changes made for this release, see ChangeLogP570.txt for filenames or keywords of interest and change summaries, or go directly to the PAPI git repository.
- Added the component PCP (Performance Co-Pilot, IBM) which allows access toPCP events via the PAPI interface.
- Added support for IBM POWER9 processors.
- Added power monitoring support for AMD Fam17h architectures via RAPL.
- Added power capping support for NVIDIA GPUs.
- Added benchmarks and testing for the “nvml” component, which allowspower-management (reporting and setting) for NVIDIA GPUs.
- Re-implementation of the “cuda” component to better handle GPU events,metrics (values computed from multiple events), and NVLink events, each ofwhich have differently handling requirements and may require separate readgroupings.
- Enhanced NVLink support, and added additional tests and example code forNVLink (high-speed GPU interconnect).
- Extension of test suite with more advanced testing: attach_cpu_sys_validate,attach_cpu_validate, event_destroy test, openmp.F test, attach_validate test(rdpmc issue).
- ARM64 configuration now works with newer Linux kernels (>=3.19).
- As part of the “cuda” component, expanded CUPTI-only tests to distinguishbetween PAPI or non-PAPI issues with NVIDIA events and metrics.
- Many memory leaks have been corrected. Not all, some 3rd party librarycodes still exhibit memory leaks.
- Better reporting and error handling of bugs. Changes to “infiniband_umad”name reporting to distinguish it from the “infiniband” component.
- Cleaning up of the source code, added documentation and test/utility files.
Acknowledgements: This release is the result of efforts from many people. The PAPI team would like to express special Thanks to Vince Weaver, Stephane Eranian (for libpfm4), William Cohen, Steve Kaufmann, Phil Mucci, and Konstantin Stefanov.