DLA: Compiler and FPGA Overlay for Neural Network Inference Acceleration

TL;DR

This paper presents a specialized FPGA overlay and a domain-specific compiler for neural network inference, achieving high performance with minimal overhead and enabling fast development cycles for deep learning applications.

Contribution

It introduces a low-overhead, application-specific FPGA overlay and a graph compiler that optimizes deep learning models for efficient inference.

Findings

Achieves ~1% overhead for overlay control and reprogramming.

Provides architecture-driven optimizations boosting CNN/RNN performance.

Reaches ~900 fps on GoogLeNet, the fastest reported on similar FPGAs.

Abstract

Overlays have shown significant promise for field-programmable gate-arrays (FPGAs) as they allow for fast development cycles and remove many of the challenges of the traditional FPGA hardware design flow. However, this often comes with a significant performance burden resulting in very little adoption of overlays for practical applications. In this paper, we tailor an overlay to a specific application domain, and we show how we maintain its full programmability without paying for the performance overhead traditionally associated with overlays. Specifically, we introduce an overlay targeted for deep neural network inference with only ~1% overhead to support the control and reprogramming logic using a lightweight very-long instruction word (VLIW) network. Additionally, we implement a sophisticated domain specific graph compiler that compiles deep learning languages such as Caffe or Tensorflow to easily target our overlay. We show how our graph compiler performs architecture-driven software optimizations to significantly boost performance of both convolutional and recurrent neural networks (CNNs/RNNs) - we demonstrate a 3x improvement on ResNet-101 and a 12x improvement for long short-term memory (LSTM) cells, compared to naive implementations. Finally, we describe how we can tailor our hardware overlay, and use our graph compiler to achieve ~900 fps on GoogLeNet on an Intel Arria 10 1150 - the fastest ever reported on comparable FPGAs.