LoopTree Tutorial

This is the tutorial repository for "LoopTree: Exploring the Fused-layer Dataflow Accelerator Design Space" by Michael Gilbert, Yannan Nellie Wu, Joel S. Emer, and Vivienne Sze.

The tutorial can be easily setup using Docker by following the steps described below.

Quick Start

Please make a copy of the docker-compose.template.yaml file named docker-compose.yaml and update the file as instructed (instructions are within the template).

Then, run the following command.

docker compose up

More Resources

You might find the following resources useful.

• The LoopTree paper.
• Timeloop/Accelergy v4 documentation.
• The PyTimeloop code.
• The Timeloop code.

Citation

If you use LoopTree in your research, please cite the LoopTree paper.

M. Gilbert, Y. N. Wu, J. S. Emer and V. Sze, "LoopTree: Exploring the Fused-layer Dataflow Accelerator Design Space," in IEEE Transactions on Circuits and Systems for Artificial Intelligence, doi: 10.1109/TCASAI.2024.3461716. keywords: {System-on-chip;Filters;Tensors;Space exploration;Taxonomy;Shape;Energy consumption},

Or, use the following BibTex.

@ARTICLE{looptree,
  author   = {Gilbert, Michael and Wu, Yannan Nellie and Emer, Joel S. and Sze, Vivienne},
  journal  = {IEEE Transactions on Circuits and Systems for Artificial Intelligence}, 
  title    = {LoopTree: Exploring the Fused-layer Dataflow Accelerator Design Space}, 
  year     = {2024},
  volume   = {},
  number   = {},
  pages    = {1-15},
  keywords = {System-on-chip;Filters;Tensors;Space exploration;Taxonomy;Shape;Energy consumption},
  doi      = {10.1109/TCASAI.2024.3461716}}