Skip to content

Latest commit

 

History

History
77 lines (43 loc) · 2.21 KB

README.md

File metadata and controls

77 lines (43 loc) · 2.21 KB
Raw

BARVINN: A Barrel RISC-V Neural Network Accelerator:

alt text

How to Use:

You first need to clone the repository and update the submodules:

git clone https://github.com/hossein1387/BARVINN.git
cd BARVINN
git submodule update --init

Now that you cloned the BARVINN repository, you can run a sample code. First make sure the Vivado is sourced, example for Vivado 2023.1:

source /opt/Xilinx/Vivado/2023.1/settings64.sh

Then make sure you have fusesoc installed:

python3 -m pip install fusesoc

Then add mvu, pito and barvinn to your fusesoc libraries (NOTE: if you have used FuseSoC with any of the following projects, you can skip adding it to FuseSoC) :

fusesoc library add barvinn .

Then run simulation (No GUI):

fusesoc run --target=sim barvinn

For synthesis:

fusesoc run --target=synth barvinn

To open sim in GUI mode:

cd build/pito_0/sim-vivado/ 
make run-gui

And for synthesis:

cd build/pito_0/synth-vivado/ 
make build-gui

This should open the project for you. Make sure you have run simulation or synthesis atleast once, otherwise fusesoc would not create a project file for you.

Documentation:

BARVINN documentation is available in docs/ folder. However, you can follow this url for an online version of documentation hosted on readthedocs. This url will ocassionally gets updated. You can build the lates docs using the following:

cd docs
python pip -r install requirements
make html

Then, you can open ./docs/_build/html/index.html file.

Publications

If you liked this project, please consider citing our paper:

@Article{barvinn_aspdac,
  author={AskariHemmat, MohammadHossein and Bilaniuk, Olexa and Wagner, Sean and Hariri, Yassine and Savaria, Yvon and David, Jean-Pierre},
  journal= {28th Asia and South Pacific Design Automation Conference ASP-DAC 2023},
  title  = {BARVINN: Arbitrary Precision DNN Accelerator Controlled by a RISC-V CPU},
  year   = {2023},
  doi    = {10.1145/3566097.3567872}
}

Link to our paper on arXiv.