Skip to content

HeteroCL: A Multi-Paradigm Programming Infrastructure for Software-Defined Heterogeneous Computing

License

Notifications You must be signed in to change notification settings

ybai62868/heterocl

 
 

Repository files navigation

GitHub license CircleCI

HeteroCL: A Multi-Paradigm Programming Infrastructure for Software-Defined Reconfigurable Computing

Website | Installation | Tutorials | Samples | Documentation

Introduction

With the pursuit of improving compute performance under strict power constraints, there is an increasing need for deploying applications to heterogeneous hardware architectures with accelerators, such as GPUs and FPGAs. However, although these heterogeneous computing platforms are becoming widely available, they are very difficult to program especially with FPGAs. As a result, the use of such platforms has been limited to a small subset of programmers with specialized hardware knowledge.

To tackle this challenge, we introduce HeteroCL, a programming infrastructure comprised of a Python-based domain-specific language (DSL) and a compilation flow. The HeteroCL DSL provides a clean programming abstraction that decouples algorithm specification from three important types of hardware customization in compute, data types, and memory architectures. HeteroCL can further capture the interdependence among these different customization techniques, allowing programmers to explore various performance/area/accuracy trade-offs in a systematic and productive manner. In addition, our framework currently provides two advanced domain-specific optimizations with stencil analysis and systolic array generation, which produce highly efficient microarchitectures for accelerating popular workloads from image processing and deep learning domains.

Current Compilation Flow

flow

Evaluation on AWS F1 (Xilinx Virtex UltraScale+TM VU9P FPGA)

The speedup is over a single-core single-thread CPU execution on AWS F1.

Benchmark Data Sizes & Type #LUTs #FFs #BRAMs #DSPs Freqency (MHz) Speedup Back End
KNN Digit Recognition
Image classification
K=3 #images=1800
uint49
4009 5835 88 0 250 12.5 General
K-Means
Clustering
K=16 #elem=320 x 32
int32
212708 235011 32 1536 190.6 16.0 General
Smith-Waterman
Genomic sequencing
string len=128
uint2
110841 88369 1409 0 152.2 20.9 General
Seidel
Image processing
2160 pixel x 3840 pixel
fixed16
21719 31663 46 96 250 5.9 Stencil
Gaussian
Image processing
2160 pixel x 3840 pixel
fixed16
70833 131160 46 688 250 13.2 Stencil
Jacobi
Linear algebra
2160 pixel x 3840 pixel
fixed16
14883 22485 46 48 250 5.0 Stencil
GEMM
Matrix multiplication
1024 x 1024 x 1024
fixed16
454492 800283 932 2507 236.8 8.9 Systolic Array
LeNet Inference
CNN
MNIST
fixed16
362291 660186 739.5 1368 250 10.6 Systolic Array

Publication

If you use HeteroCL in your design, please cite our FPGA'19 paper:

@article{lai2019heterocl,
  title={HeteroCL: A Multi-Paradigm Programming Infrastructure for Software-Defined Reconfigurable Computing},
  author={Lai, Yi-Hsiang and Chi, Yuze and Hu, Yuwei and Wang, Jie and Yu, Cody Hao and 
          Zhou, Yuan and Cong, Jason and Zhang, Zhiru},
  journal={Int'l Symp. on Field-Programmable Gate Arrays (FPGA)},
  year={2019}
}

Related Work

HeteroCL is a Python-based DSL extended from TVM and it extends Halide IR for intermediate representation. HeterCL incoporates the SODA framework, PolySA framework, and Merlin Compiler for FPGA back-end generation.

Contributing to HeteroCL

  1. Use Pull Request.
  2. Python coding style.
  3. Python docstring style.

About

HeteroCL: A Multi-Paradigm Programming Infrastructure for Software-Defined Heterogeneous Computing

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C++ 76.7%
  • Python 19.6%
  • Java 0.9%
  • Makefile 0.8%
  • C 0.6%
  • CMake 0.6%
  • Other 0.8%