🚀 A High-Performance FPGA Placement and Routing Framework Powered by PyTorch

State-of-the-Art  |  GPU-Accelerated  |  Production-Ready

Table of Contents

• Overview
• Key Features
• Latest Updates
• Visual Demo
• Getting Started
  • Prerequisites
  • Installation
  • Running Benchmarks
• Documentation
• Resources
• Team & Contributors
• Publications
• License

Overview

OpenPARF is a cutting-edge FPGA placement and routing framework that delivers superior performance through innovative algorithms and GPU acceleration. Our framework achieves up to 12.7% improvement in routed wirelength and 2X+ speedup in placement compared to state-of-the-art solutions.

Key Features

• 🔋 Advanced Placement Technology

  • Multi-Electrostatic-based placement engine
  • Multi-die FPGA support with SLL count optimization
  • Comprehensive unit support (LUT, FF, DSP, BRAM, IO, etc.)

• ⚡ High Performance

  • GPU acceleration via CUDA
  • 0.4-12.7% better routed wirelength
  • 2X+ faster placement speed

• 🛠 Production Ready

  • Industry-standard architecture support
  • Extensive benchmarking suite
  • Well-documented API and examples

Latest Updates

🎉 OpenPARF 2.0 Release (2024/02/02)

• Multi-die FPGA placement support
• Flexshelf architecture format
• Performance improvements and bug fixes
• Release Notes

Visual Demo

Experience the power of our electrostatic field-based placement through these visualizations from benchmark ISPD2016/FPGA06:

LUT	FF	DSP	BRAM

Getting Started

Prerequisites

• Python 3.7+
• C++14 compatible compiler
• PyTorch 1.7.1
• CUDA 11.0 (for GPU acceleration)
• Gurobi 9.5 (optional)

Installation

# Create conda environment
mamba create --name openparf python=3.7
mamba activate openparf

# Install dependencies
mamba install cmake boost bison
mamba install pytorch==1.7.1 torchvision==0.8.2 cudatoolkit=11.0 -c pytorch
pip install hummingbird-ml pyyaml networkx tqdm

# Build OpenPARF
git clone --recursive https://github.com/PKU-IDEA/OpenPARF.git
mkdir build && cd build
cmake ../OpenPARF -DCMAKE_PREFIX_PATH=$CONDA_PREFIX -DPYTHON_EXECUTABLE=$(which python)
make -j8 && make install

For detailed installation instructions, see INSTALL.md.

Running Benchmarks

cd <installation directory>
python openparf.py --config unittest/regression/ispd2016/FPGA01.json

For more benchmark information, see BENCHMARKS.md.

Documentation

• 📘 Installation Guide
• 📗 Benchmark Guide
• 📙 Architecture Customization

Resources

• ISPD 2016 Contest
• ISPD 2017 Contest

Team & Contributors

OpenPARF is developed and maintained by PKU-IDEA Lab at Peking University, under the supervision of Prof. Yibo Lin.

Core Contributors:

• Jing Mai, Jiarui Wang and Yibai Meng - Initial release
• Runzhe Tao - Multi-die placement integration
• Jing Mai, Jiarui Wang, Yifan Chen, Zizheng Guo, Xun Jiang, Yun Liang, Yibo Lin - OpenPARF 3.0 with macro placement

Publications

For academic papers and citations, please see our publications list.

License

OpenPARF is released under the BSD 3-Clause License. See LICENSE for details.