A Version of Parallel Odd-Even Sorting Algorithm Implemented in CUDA Paradigm

Sorting data is an important problem for many applications. Parallel sorting is a way to improve sorting performance using more nodes or threads e.g. dividing data in more nodes and perform sorting in each node simultaneously or including more threads in process of sorting. It was experimented with one type of those sorting algorithms, namely the well-known sorting algorithms called Odd-Even sort. This paper describes a modification of the above mentioned algorithm. Namely, the algorithm modification consists in the ability to work with the blocks of elements instead of working with individual elements. This modification is done with the idea to make it in a closer form for use of the CUDA technology. Both theoretical and experimental analysis of Odd-Even sort algorithm together with its parallel implementation is done. For experimental purpose, a GeForce GT 645M with 2 GB memory is used. The programming language C++ with CUDA 7.0 paradigm is utilized to implement Odd-Even algorithm and the results indicated that sorting of integers in CUDA environment are dozens of times faster.