Version 1.0.0b3

This is a beta release for early access to the new features. Not intended for production use.

This version contains several bugfixes:

• Fix snap_to_nearest on non-numeric parameters by @stijnh in #221
• Fixed an issue where some restrictions would not be recognized by the old check_restrictions function.
• Fixed an issue where bayes_opt would not handle pruned parameters correctly.

Full Changelog: 1.0.0b2...1.0.0b3

Version 1.0.0b2

This is a beta release for early access to the new features. Not intended for production use.

Full Changelog: 1.0.0b1...1.0.0b2

Version 1.0.0 beta 1

This is a beta release for early access to the new features. Not intended for production use.

What's Changed

• HIP Backend by @MiloLurati in #199
• Accuracy tuning by @stijnh in #189
• Fix issue where HIP backend fails due to invalid arguments type by @stijnh in #216
• Searchspace improvements and project meta modernization by @fjwillemsen in #214
• Minor bugfix by @isazi in #219
• OpenACC support by @isazi in #197
• Fixed broken tests as per issue #217 by @fjwillemsen in #220

New Contributors

• @MiloLurati made their first contribution in #199

Full Changelog: 0.4.5...1.0.0b1

Version 0.4.5

Version 0.4.5 adds support of using PMT in combination with Kernel Tuner enabling power and energy measurements on a wide range of devices. In addition, we have worked extensively on the internals of Kernel Tuner and the interfaces of the separate components that together make up Kernel Tuner. Along with a few bugfixes, fixes of small errors in examples and documentation.

[0.4.5] - 2023-06-01

Added

• PMTObserver to measure power and energy on various platforms

Changed

• Improved functionality for storing output and metadata files
• Updated PowerSensorObserver to support PowerSensor3
• Refactored interal interfaces of runners and backends
• Bugfix in interface to set objective and optimization direction

Version 0.4.4

Version 0.4.4

Version 0.4.4 adds extended support for energy efficiency tuning. In particular, with the new capability to fit a performance model to the target GPUs power-frequency curve. How to use these features is demonstrated in:
https://github.com/KernelTuner/kernel_tuner/blob/master/examples/cuda/going_green_performance_model.py

And described in the paper:

Going green: optimizing GPUs for energy efficiency through model-steered auto-tuning
R. Schoonhoven, B. Veenboer, B. van Werkhoven, K. J. Batenburg
International Workshop on Performance Modeling, Benchmarking and Simulation of High Performance Computer Systems (PMBS) at Supercomputing (SC22) 2022
https://arxiv.org/abs/2211.07260

Other than that, we've implemented a new output and metadata JSON format that adheres to the 'T4' auto-tuning schema created by the auto-tuning community at the Lorentz Center workshop in March 2022.

From the changelog:

[0.4.4] - 2023-03-09

Added

• Support for using time_limit in simulation mode
• Helper functions for energy tuning
• Example to show ridge frequency and power-frequency model
• Functions to store tuning output and metadata

Changed

• Changed what timings are stored in cache files
• No longer inserting partial loop unrolling factor of 0 in CUDA

Version 0.4.3

The version 0.4.3 release consists of a large number of changes to the internals of Kernel Tuner, including the addition of a new backend based on Nvidia's official Python bindings for CUDA, as well as improved functionality for tuning energy efficiency, e.g. measuring core voltages, the measurement of power and the interface with NVML has also improved a lot.

Some of the changes are also in the "externals" of Kernel Tuner. In the sense that we have migrated from https://github.com/benvanwerkhoven/ to https://github.com/KernelTuner. The goal of this move is to bring the collection of repositories belonging to the larger Kernel Tuner project under one organization.

From the Changelog:

[0.4.3] - 2022-10-19

Added

• A new backend that uses Nvidia cuda-python
• Support for locked clocks in NVMLObserver
• Support for measuring core voltages using NVML
• Support for custom preprocessor definitions
• Support for boolean scalar arguments in PyCUDA backend

Changed

• Migrated from github.com/benvanwerkhoven to github.com/KernelTuner
• Significant update to the documentation pages
• Unified benchmarking loops across backends
• Backends are no longer context managers
• Replaced the method for measuring power consumption using NVML
• Improved NVML measurements of temperature and clock frequencies
• bugfix in parse_restrictions when using and/or in expressions
• bugfix in GreedyILS when using neighbor method "adjacent"
• bugfix in Bayesian Optimization for small problems

Version 0.4.2

Version 0.4.2 includes a lot of work on the search space representation, application of restrictions, and optimization strategies. In addition to the addition of several new optimization strategies, most optimization strategies should see improved performance both in terms of the number of evaluated kernel configurations as well as execution time.

Added

• new optimization strategies: dual annealing, greedly ILS, ordered greedy MLS, greedy MLS
• support for constant memory in cupy backend
• constraint solver to cut down time spent in creating search spaces
• support for custom tuning objectives
• support for max_fevals and time_limit in strategy_options of all strategies

Removed

• alternative Bayesian Optimization strategies that could not be used directly
• C++ wrapper module that was too specific and hardly used

Changed

• string-based restrictions are compiled into functions for improved performance
• genetic algorithm, MLS, ILS, random, and simulated annealing use new search space object
• diff evo, firefly, PSO are initialized using population of all valid configurations
• all strategies except brute_force strictly adhere to max_fevals and time_limit
• simulated annealing adapts annealing schedule to max_fevals if supplied
• minimize, basinhopping, and dual annealing start from a random valid config

Version 0.4.1

This version adds a brand new Bayesian Optimization strategy, as well as some smaller features and fixes.

[0.4.1] - 2021-09-10

Added

• support for PyTorch Tensors as input data type for kernels
• support for smem_args in run_kernel
• support for (lambda) function and string for dynamic shared memory size
• a new Bayesian Optimization strategy

Changed

• optionally store the kernel_string with store_results
• improved reporting of skipped configurations

Version 0.4.0

This version adds a great deal of new functionality and extra flexibility and additional control to the user over what is being benchmarked and when. From the CHANGELOG:

Added

• support for (lambda) function instead of list of strings for restrictions
• support for (lambda) function instead of list for specifying grid divisors
• support for (lambda) function instead of tuple for specifying problem_size
• function to store the top tuning results
• function to create header file with device targets from stored results
• support for using tuning results in PythonKernel
• option to control measurements using observers
• support for NVML tunable parameters
• option to simulate auto-tuning searches from existing cache files
• Cupy backend to support C++ templated CUDA kernels
• support for templated CUDA kernels using PyCUDA backend
• documentation on tunable parameter vocabulary

Version 0.3.2

Version 0.3.2

This version adds several new and recent features. Most importantly is the new feature to specify user-defined metrics for Kernel Tuner to compute along with the benchmarking results. User-defined metrics are composable, so you can define metrics that build upon other metrics. The documentation pages have also been updated to include this new feature and other recent changes.

An important change that might influence benchmark results reported by Kernel Tuner is the fact that the runner will now do a warm up of the device using the first kernel in the parameter space. This is to remove any startup or cold start delays that were significantly slowing down the first benchmarked kernel on many devices.

From the changelog:

[0.3.2] - 2020-11-04

Added

• support loop unrolling using params that start with loop_unroll_factor
• always insert "define kernel_tuner 1" to allow preprocessor ifdef kernel_tuner
• support for user-defined metrics
• support for choosing the optimization starting point x0 for most strategies

Changed

• more compact output is printed to the terminal
• sequential runner runs first kernel in the parameter space to warm up device
• updated tutorials to demonstrate use of user-defined metrics