high performance computing on graphics processing units: hgpu.org

Graphs are a popular data structure to represent large amounts of data and the relationship between them. As serial hardware hits the wall in terms of computation speed, a lot of research has been made recently in parallelizing Graph Search Algorithms such as Breadth First Search or the Single Source Shortest Path Problem hence make them applicable to manycore processors and accelerators such as GPUs. OpenCL offers a framework to write task- and data-parallel programs which execute across heterogenous platforms, thus should be suitable when implementing parallel Algorithms for Graphs. This thesis explains how Breadth First Search, Single Source Shortest Path Problem, All Pair Shortest Path Problem, Transposition of Graph and creating a Topological Order for acyclic Graphs can be parallelized efficiently by looking at the the current research. In addition, a Library with different strategies for each algorithm was implemented in OpenCL. Each Implementation was tested with random generated Graphs on a Intel i5-3570k CPU and a AMD Radeon HD 7900 GPU and their execution time compared. The results showed that GPUs are suitable for processing Graph Search Algorithms.