This repository provides access to the Sunstone scheduler, a tool for mapping tensor computations on the PE array - style accelerators with explictly managed memories. The design principles behind the Sunstone is discussed in our research paper Sunstone: A Scalable and Versatile Scheduler for Mapping Tensor Algebra on Spatial Accelerators, published in ISPASS 2023.
- Efficient mapping of tensor computations on PE array-style accelerators
- Explicit management of memories to minimize data movement
- Scalable scheduling algorithm that is fast even for modern multi-level architectures
To get started with the Sunstone scheduler, follow these steps:
- Clone this repository
git clone https://github.com/compstruct/sunstone.git
- create a vritual environment
cd sunstone
python3 -m venv venv
source venv/bin/activate
- Install the required dependencies
pip install -r requirements.txt
- Run the examples to make sure everything works
python examples/eyeriss_like/main.py
We strongly encourage new users to look into the examples directory and gain more familiarity with Sunstone.
We welcome contributions from the community. If you have any suggestions or find any bugs, please open an issue or submit a pull request.
The Sunstone scheduler is licensed under the BSD 2-Clause License.