General Parallel Computation on Commodity Graphics Hardware: Case Study with the All-Pairs Shortest Paths Problem

摘要

Programmability and IEEE-standard floating point arithmetic makes the latest commodity graphics processors (GPUs) an attractive platform for general parallel computing. In this paper -we describe the implementation of the Warshall-Floyd algorithm on a class of GPUs. All-pairs shortest paths problem is relevant to many practical applications. Efficient GPU implementation of the Warshall-Floyd algorithm is challenging due to the algorithm's dynamic nature as well as limited GPU instruction set. GPU specific data organization, parallelization, and experimental results for several graphics accelerators are discussed. Algorithm implementation on the GPU utilizes interpolators, vertex and fragment pipelines, as well as vector operations to maximize performance. Speedups of up to 3x over a CPU implementation were achieved.