Home // ICSEA 2019, The Fourteenth International Conference on Software Engineering Advances // View article

OpenCL-Generated Optimizing Compiler for FPGA Using ROSE Compiler Infrastructure

Authors:
Yuichiro Aoki

Keywords: FPGA, OpenCL, Compiler, Parallel Programming

Abstract:
Many researchers are investigating deep learning because it can recognize pedestrians for automatic driving and/or criminals to prevent crimes on the street. A promising device for such tasks in deep learning is a Field Programmable Gate Array (FPGA). However, the conventional manual FPGA programming and optimizations are complicated and take a long time. Thus, FPGA development time needs to be decreased. In this paper, we propose an OpenCL-generated optimizing compiler based on the ROSE Compiler Infrastructure. OpenCL is a C-extended programming language for heterogeneous computing, such as an FPGA and a Central Processing Unit (CPU). We add simple pragmas to the C program, and our compiler generates the optimized OpenCL program for FPGA. The preliminary evaluation using the deep learning framework Caffe shows that our compiler decreases to about 1/16 of the conventional development time.

Pages: 57 to 60

Copyright: Copyright (c) IARIA, 2019

Publication date: November 24, 2019

Published in: conference

ISSN: 2308-4235

ISBN: 978-1-61208-752-8

Location: Valencia, Spain

Dates: from November 24, 2019 to November 28, 2019