An investigation into the performance of general sorting on graphics processing units

Abstract

Sorting is a fundamental operation in computing and there is a constant need to push the boundaries of performance with different sorting algorithms. With the advent of the programmable graphics pipeline, the parallel nature of graphics processing units has been exposed allowing programmers to take advantage of it. By transforming the way that data is represented and operated on parallel sorting algorithms can be implemented on graphics processing units where previously only graphics processing could be performed. This paradigm of programming exhibits potentially large speedups for algorithms. © Springer Science+Business Media B.V. 2008.