New GPU Sorting Algorithm Using Sorted Matrix

Abstract

The sorting of elements is vital in various real-life applications such as CPU scheduling in operating systems, recommendation systems based on search time, Television channels sorted based on view time, etc. But it becomes complex and time-consuming if the number of elements becomes larger. For this problem, multiple solutions have been proposed in the past; among them, one efficient solution uses the concept of finding the kth element in a column and row-wise sorted matrix. If a sorted matrix is used to represent a list's items, it can be used to determine any element's position in the sorted list. In this paper, a Graphics Processing Unit (GPU) based parallel method has been proposed to sort a list using the concept of the sorted matrix. A mathematical proof of the proposed method and its implementation for the GPU-based machine is also presented. Experimental results of the proposed algorithm are compared with the CUDA Thrust Library Radix sorting implementation. The proposed algorithm has achieved a five-fold speed increase compared to the CUDA Thrust library sorting performance.