Matmap: A Modular, Automatable, Tunable Mapper for Accelerator Programming

A representation for higher-level transforms, with tools to generate EXO code.

Setup

This has been extensively tested on Python 3.9.7. Python versions 3.7 and earlier are not supported as Exo requires several newer language features not available.

git clone https://github.com/gdinh/matmap.git
cd MoST
python -m venv $HOME/.venv/matmap
source $HOME/.venv/matmap/bin/activate
python -m pip install pip setuptools wheel
python -m pip install -r requirements.txt

If you'd like to run the interactive demos/documentation, install iPython and set it up so that it points to your venv by running these from within the venv:

python -m pip install jupyter
ipython kernel install --user --name=matmap_venv

Now, you'll be able to launch jupyter notebook; just navigate to docs/ and open most_demo.ipynb, then point the notebook to the venv using Kernel > Change kernel > matmap_venv.

Project status:

• Tiling schedule works
• HBL projective optimal tiling works for constant loop bounds

In progress:

• GPTune integration
• CoSA transform

Next TODOs:

• HBL autotiling for CNNs
• Code specialization for variable sized bounds