GasCL: A vertex-centric graph model for GPUs

Author

Shuai Che

Author_Institution

Adv. Micro Devices, USA

fYear

2014

fDate

9-11 Sept. 2014

Firstpage

1

Lastpage

6

Abstract

There are increasing research efforts of using GPUs for graph processing. Most prior work on accelerating GPGPU graph algorithms has been focused on algorithm and device-specific optimizations. There is little research on studying high-level programming models and associate run-time systems for graph processing on GPUs, which will be useful to solve diverse real-world problems flexibly. This paper presents a preliminary implementation of a graph framework, GasCL, supporting the well-known “think-like-a-vertex” programming model. The system is built on top of OpenCL and portable across diverse accelerators. We describe our design and use two applications as case studies. The initial performance result shows an average of 2.5× speedup on a GPU compared with a CPU.

Keywords

graph theory; graphics processing units; optimisation; parallel programming; GPGPU graph algorithms; GasCL; OpenCL; device-specific optimizations; gather-apply-scatter; general purpose graphics processing unit; graph framework; graph processing; think-like-a-vertex programming model; vertex-centric graph model; Arrays; Computational modeling; Graphics processing units; Kernel; Programming; Runtime;

fLanguage

English

Publisher

ieee

Conference_Titel

High Performance Extreme Computing Conference (HPEC), 2014 IEEE

Conference_Location

Waltham, MA

Print_ISBN

978-1-4799-6232-7

Type

conf

DOI

10.1109/HPEC.2014.7040962

Filename

7040962