Mapping the MPM maximum flow algorithm on GPUs

Abstract

The GPU offers a high degree of parallelism and computational power that developers can exploit for general purpose parallel applications. As a result, a significant level of interest has been directed towards GPUs in recent years. Regular applications, however, have traditionally been the focus of work on the GPU. Only very recently has there been a growing number of works exploring the potential of irregular applications on the GPU. We present a work that investigates the feasibility of Malhotra, Pramodh Kumar and Maheshwari's "MPM" maximum flow algorithm on the GPU that achieves an average speedup of 8 when compared to a sequential CPU implementation. © 2010 IOP Publishing Ltd.