A Practical Quicksort Algorithm for Graphics Processors

Abstract

Abstract. In this paper we present GPU-Quicksort, an efficient Quicksort algorithm suitable for highly parallel multi-core graphics processors. Quicksort has previously been considered as an inefficient sorting solution for graphics processors, but we show that GPU-Quicksort often performs better than the fastest known sorting implementations for graphics processors, such as radix and bitonic sort. Quicksort can thus be seen as a viable alternative for sorting large quantities of data on graphics processors.

K-Model: A New Computational Model for Stream Processors

“…However, it is far from being accurate: for example it does not consider the cost of nonlocal memory references that, instead, has a great impact on performance. The PRAM model has been applied to some recently proposed GPU algorithms [4]. Indeed, as we describe in this paper, experimental results contradict theoretical results that are not done using a computational model specifically designed for GPUs.…”

GPU Matrix Sort (An Efficient Implementation of Merge Sort)

“…This means, however, that there will be a very little parallelization in the beginning, when the sequence is few [6]. To deal with this issue, Caderman [7] introduces a parallel version of quick sort combining CPU process and GPU process. The algorithm is theoretically optimal, but the data transferring between the CPU and GPU slows down the running time in practice.…”

Fast Parallel Sorting Algorithms on GPUs

“…An overview of sorting algorithms in parallel is given in [8]. A quick-sort implementation on GPU using CUDA is considered in [9] which results quick-sort as an efficient alternative to both bitonic and radix sort over GPU's for larger data sequences. Moreover bitonic sort is suggested for smaller sequences.…”

“…Moreover bitonic sort is suggested for smaller sequences. The quick-sort algorithm discussed in [9] uses a divide-and-conquer approach for sorting, forming left and right sequences depending on whether current value is greater or smaller than pivot value. For each recursive call, a new pivot value has to be selected.…”

“…For each recursive call, a new pivot value has to be selected. On the GPU, [9] have proposed two steps (1) sub-sequences creation, and (2) assigning each sub-sequence to thread for sorting. The overall complexity of above GPU-quick sort technique is ) log ( n n Ο A merge and radix-sort implementation for GPU's is provided in [10].…”