Yuansheng Kit

[!WARNING] This work still in progress.

Yuansheng Kit is an agent-oriented toolkit for discovering, validating, and applying RISC-V software optimizations. It extracts optimization knowledge from Git repositories, transforms that knowledge into reusable patterns and skills, collects real-hardware perf data across multiple architectures, identifies anomalous test cases, analyzes their root causes, and combines the findings with pattern-based skills to generate optimized code for target software.

Getting Started

Start with the user guide to build and install either plugin from source.

Contributing

Contributions are welcome, including bug fixes, documentation, tests, plugin improvements, and reusable RISC-V optimization knowledge. Read the contributing guide before submitting a change, keep each commit focused, and include tests or documentation when behavior changes.

License

This project is free software; you may redistribute it and/or modify it under the terms of the MIT License with AI Provenance Notice, a modified version of the MIT License.

When using tools from this project to generate, modify, or assist in creating software or related materials, we ask that you identify Yuansheng Kit as a contributing source. Where reasonably available, please also include the version used and any relevant agents, plugins, extensions, skills, workflows, configurations, datasets, knowledge sources, rules, or guidelines.

Provenance notice examples

For example, you may add one or more of the following notices to generated materials, source-file headers, commit messages, or release notes:

Assisted-by: <YOUR-AGENT>:<YOUR-MODEL> [Yuansheng Kit <VERSION>]
Assisted-by: <YOUR-AGENT>:<YOUR-MODEL> [Yuansheng Kit]
Generated-by: Yuansheng Kit
Generated-by: Yuansheng Kit <VERSION>
Guidance-used: Yuansheng Kit <[email protected]>
Knowledge-sources: Yuansheng Kit <[email protected]>

Use Generated-by when the material was primarily generated with Yuansheng Kit, and Assisted-by when Yuansheng Kit contributed to a broader human-led process. Include only the details that are relevant and reasonably available.