SASA

SASA is a scalable and automatic stencil acceleration framework on modern HBM-based FPGAs. It automatically parses a stencil DSL, exploits the best hybrid spatial and temporal parallelism configuration on an HBM-based FPGA, and generates the optimal FPGA design in Vitis HLS with TAPA/AutoBridge-based floorplanning optimization. If you use SASA in your research, please cite our paper:

Xingyu Tian, Zhifan Ye, Alec Lu, Licheng Guo, Yuze Chi, and Zhenman Fang. 2022. SASA: A Scalable and Automatic Stencil Acceleration Framework for Optimized Hybrid Spatial and Temporal Parallelism on HBM-based FPGAs. Accepted by ACM Transactions on Reconfigurable Technology and Systems (TRETS 2023).

Download

git clone https://github.com/SFU-HiAccel/SASA.git

Setup Requirement

1. Evaluated hardware platforms:
   • Host OS
     • 64-bit Ubuntu 18.04.2 LTS
   • Datacenter FPGA
     • Xilinx Alveo U280 - HBM2-based FPGA
2. Software tools:
   • HLS tool:
     • Vitis 2020.2, 2021.1 or 2021.2
     • Xilinx Runtime (XRT) corresponding version with Vitis
   • Python 3.6+
   • TAPA

SASA DLS Example (stencil.dsl)

# Kernel name
kernel: JACOBI2D

# Required iteration
iteration: 4

# Input name and size
input float: in_1(9720, 1024)
# Stencil kernel pattern with relative coordinates
output float: out_1(0, 0) = ( in_1(0,1) + in_1(1,0) + in_1(0,0) + in_1(0,−1) + in_1(−1,0) ) / 5

# Optional parameters can be decided automatically or user can specify congfiguration
# Kernel number
count: 3
# PE number per kernel
repeat: 4
# Parallelism
boarder: streaming

Usage

SASA takes high-level DSL description as inputs, automatically explores the best spatial and temporal parallelism.

1. Generate stencil design with the best optimization:

   • Run: python3 exploration.py --src stencil.dsl

2. Generate customized stencil design

   • Generate the both host code and kenrel code in Vitis HLS
     • Run: python3 codgen.py --src stencil.dsl
   • Process HLS synthesis
     • Run: source generate_xo.sh
   • Generate bitstream
     • Run: source generate_bitstream.sh
   • Compile host code
     • Run: g++ -o stencil -O2 stencil.cpp host.cpp -ltapa -lfrt -lglog -lgflags -lOpenCL

3. Execute on hardware

   • Run: ./stencil --bitstream=stencil.$platform.hw.xclbin
   • E.g. $platform = xilinx_u280_xdma_201920_3

Contact

• Xingyu Tian, PhD Student
• HiAccel Lab, Simon Fraser University (SFU)
• Supervisor: Dr. Zhenman Fang
• Email: [email protected]