Variorum 0.8.0

Version 0.8.0 is a major release.

New features

• Adds energy APIs: Print and JSON (#530)
• Update powmon to report utilization in addition to power. (#510)
• Update JSON utilization API to support CPU-only, GPU-only and multi-architecture
  systems (#525)
• Support GPUs and multi-architecture builds with new nested variorum_get_power_json API
• Adds GPU and node utilization JSON APIs (CPU, GPU, Memory) (#431)
• Add json output for frequency data (#488)
• Add json output for thermal data (#461)
• Adds libjustify as an optional build dependency for formatting tabular output (#468)

Documentation updates

• Update supported architectures (#482)
• Fix reference to Examples.rst from api/print_functions.rst

Bug fixes

• Fix Intel DRAM overflow issue (#328)
• Fix memory leak in Nvidia port
• Decref json object if unable to access ibm OCC sensor file (#477)
• Fix extern linkage issues in nvidia_gpu_power_features.h in newer versions of gcc
  (#464)

Internal changes

• Rename powmon to var_monitor throughout (#523)
• Update Variorum AMD GPU port to support ROCm 6.0.2 (#512)
• Code cleanup: change uint64_t assignments from -1 to 0 (#519)
• Update the name of variorum_get_thermals_example to match the API (#509)
• Top-level API: rename variorum_get_node_frequency_json --> variorum_get_frequency_json
  (#50)
• Add -example to utilization json example files (#501)
• Add missing ifdefs for P_*_{CPU,GPU}_IDX variables
• Format headers, parameters on new lines (#491, #493, #534)
• Add host config for folsom, alehouse cpu+gpu, alehouse gpu-only, remove host config
  for tavern, fix host config for rhetoric and thompson (#447, #457, #458, #478, #508)
• Report GPU power limit using the correct NVML API (#446)
• Change return code for variorum_cap_socket_frequency_limit from -1 to 0
• Fix function pointer for intel model/family 06-4f (#536)
• Various updates to internal GitLab and external GitHub CI

Variorum 0.7.0

Version 0.7.0 is a major release.

New features

• Enable heterogeneous build support for two architectures
• Support for new architectures: ARM Neoverse N1 platform, Sapphire Rapids
• Add power capping API for NVIDIA, AMD and Intel GPUs
• Add get power limit for intel GPUs

Documentation updates

• add links to youtube video and recordings from ECP Tutorial 2023
• Change knob to dial in documentation
• Update Variorum Overview Figure
• Upload new logo

Bug fixes

• Fix fortran error with gfortran > 10
• Fix issues coming up with DENABLE_WARNINGS=On on AMD CPU port
• Fix AMD build errors: add missing headers, undefined variables, etc.
• Fix inconsistent spacing issue in powmon
• Fix for multiplatform build (ARM+NVIDIA) error/warnings
• Remove the undefined variable in AMD debug build

Internal changes

• Powmon rehaul
• Fix for JSON example memory leak
• Remove unnecessary call to obtain number of sockets from JSON example
• Update filenames and low-level function names, prefix power_features.* with architecture name
• Replace existing RSMI API call with the correct API to get GPU busy %
• Resolve missing non-optional arguments in control examples

Variorum 0.6.0

Version 0.6.0 is a major release.

New features

• Support for new architectures: Intel discrete GPUs, AMD GPUs through ROCm-SMI
• Add python wrappers and python module pyVariorum

API changes

• Update JSON APIs to use char* instead of json_t*
• Creates new variorum_topology.h header to expose get topology APIs

Other notable changes

• VARIORUM_LOG moved from a build variable to an environment variable
• Update architecture-related names to be more granular, for example,
  VARIORUM_WITH_AMD became VARIORUM_WITH_AMD_CPU and VARIORUM_WITH_AMD_GPU, and
  SandyBridge_2A.h became Intel_06_2A.h
• Standardized examples with -h and -v flags, for printing the help menu and
  Variorum version, respectively

Variorum 0.5.0

Version 0.5.0 is a major release.

New features

• Support for new architectures: AMD CPUs, Intel Ice Lake
• Add examples for integration Variorum into OpenMP and MPI programs, adds
  dependency on rankstr library for MPI examples (optional)

API changes

• Rename clock_speed to frequency and power_limits to power_limit throughout
  Variorum API
• Report info about power domains through JSON, such as what can be measured,
  what can be measured and capped (along with limits), and what cannot be
  measured or capped (unsupported).
• Deprecate set_and_verify API and removes associated example
• Extend JSON API and Best Effort Node Power Cap to ARM, AMD, and Intel
  platforms, IBM already supported

Other notable changes

• Use getopt to parse arguments in examples
• CI migrated from Travis to GitLab
• Add presentations from ECP 2021 tutorial series
• Extend GitLab build matrix to build across different GCC versions
• Refactor msr_core.[ch] into its own directory as it is used by both Intel and
  AMD ports
• Improved docs: add AMD and ARM docs, rankstr dependency, re-organize
  hierarchy, update system digram, add pages for categories of APIs, add info
  on Argo, PowerStack, EEHPC WG, PowerAPI, and Variorum integrations
• License header check as new unit test

Variorum 0.4.1

• Add docs for NVIDIA port
• Add units to tabular and JSON output formats
• Rename dump and print internal functions to print and print_verbose,
  respectively
• Standardize sphinx docs with rstfmt python package
• Update variorum system design figure in docs
• Speed up travis builds with parallel make

Variorum 0.4.0

• Add support for ARM Juno architecture
• Add top-level API to query available frequencies, add support for Intel and ARM
• Add JANSSON-C dependency
• Introduce JSON API to return monitoring data into a structure
• Update variorum to build and run with both msr-safe < v1.5 (msr_whitelist)
  msr-safe >= v1.5 (msr_allowlist)
• Add best effort node power cap support for Intel Broadwell and IBM
• Build option to enable all warnings, cleanup compiler warnings
• Default build type switched from Release to RelWithDebInfo
• Consistent formatting for verbose printouts across architectures
• Update year in license headers
• Update docs throughout: JSON API, best effort power cap, architectures
  supported, add sample output to examples
• Rename set functions to cap throughout top-level API (and internally in
  architecture-level functions) to clarify that the setting is an upper bound;
  instantaneous reads on these metrics may be at or below the value

Variorum 0.3.0

• Setup unit test infrastructure, test on gitlab to leverage msr-safe and OPAL
  access
• Update return code for top-level variorum APIs from -1 to 0, enables us to run unit tests on any architecture
• Fix poll_power and monitoring segfaults
• Start adding python utilities to verify system environments (i.e. msr-safe, OPAL)
• Rename variorum_set_topology to variorum_get_topology
• Rename variorum examples
• Add example integrations with cmake/make-build systems and c/cpp/fortran codes, use these in CI workflow to test install
• Use spack mirrors in CI workflow to avoid failed fetch commands
• New mailing lists: variorum-users and variorum-maintainers
• Nvidia port: use node topology to map GPU to sockets
• Fix for building variorum as static library
• Add source code (c/cpp/h/python) format check to CI workflow
• Add Contributor Covenant Code of Conduct

Variorum 0.2.0

• Add support for Nvidia platform -- Volta architecture
• Add ECP 2020 tutorial presentation
• Prefix front-facing APIs with variorum_
• Bug fix for computing power usage values
• On Intel platforms, translate both power limit time windows (not just TW1)
• Fix hwloc topology mapping (use logical indices, not OS)
• Integrate Travis and GitLab for CI
• Modify IBM powmon formatting into columns
• Update license year for 2020

Variorum 0.1.0

• Initial release to https://github.com/llnl/variorum!
• Includes support for Intel platform -- Sandy Bridge, Ivy Bridge, Haswell, Broadwell, Skylake, and Kaby Lake architectures
• Includes support for IBM platform -- Power9 architecture
• Add powmon command line sample for monitoring performance counters during application execution
• Initial RTD at https://variorum.readthedocs.io
• Rename libvariorum to variorum
• Bug fix for computing power from energy on Intel
• Add developer docs for supporting new platforms
• Add info on IBM sensors, OCC, and power shifting ratio
• Rename front-facing APIs from print_*() to print_verbose_*() and dump_*() to print_*()