Autotuning Methodology documentation

Welcome to the documentation for the autotuning_methodology package accompanying the paper “A Methodology for Comparing Auto-Tuning Optimization Algorithms”. This package makes the guidelines in the paper easy to apply: specificy the experiments json file, run autotuning_visualize [path_to_json] and wait for the results!

This is an example output of this methodology package. It depicts the aggregated performance of four optimization algorithms for two kernels on two GPUs.

Documentation Contents

• Getting Started
• Contributing
• Module contents
• Baseline module
• Caching module
• Curves module
• Experiments module
• Runner module
• Searchspace statistics module
• Visualize experiments module