English | 简体中文 | 🗃️项目开发报告

yaTuner: yet another auto tuner for compilers.

A virtual environment can be created using make init and examples are placed at directory examples.

To tune a program, use yatuner -g <filename> to generate a basic template for the program. Then modify the the script for further use. More information about the tuning script is contained in the template.

Also, yatuner.utils includes tools that might be necessary for use, here is a brief summary:

These tools can be used in the tuning script, see examples for details.

The tuning process and relative methods and their functionalities are listed as below.

Detailed documentation of yatuner.Tuner and yatuner.utils can be found in docs.

1. Install yaTuner.
2. Auto-generate tuning script with yatuner -g <filename>.
3. Manually modify tuning script and add details.
4. Run tuning script with python <filename>.

graph TB
subgraph User Interface
    O([yaTuner]) --> A
    A[(Auto-generated Tuning Script)] -- manually add details --> B[(Final Tuning Script)]
end
B == run ==> C
subgraph Tuning Process
    C(Auto-fetching compiler options) --> D(Hypothesis Test for Optimizers) 
    D --> E(Tuning Optimizers)
    E --> F(Hypothesis Test for Parameters)
    F --> G(Tuning Parameters)
    G -- run test and compare --> I[(Tuning Report)]
end
G -- store --> H[(Tuned Options)]
E -- store --> H
subgraph Tuner
    J([LinUCB Optimizer]) -.-> G
    K([Bayesian Optimizer]) -.-> G
    K -.-> E
end
subgraph Pre-defined Functions
    B == defines ==> L([comp])
    B == defines ==> M([run])
    B == defines ==> N([perf])
end
N -.-> J
L & M -.-> J & K

yaTuner is licensed under Mulan PSL v2. See LICENSE for more details.

This is a project for the OS competition 2022, proj105 problem, see this for further information.