GPU-Quicksort

Abstract

In this article, we describe GPU-Quicksort, an efficient Quicksort algorithm suitable for highly parallel multicore graphics processors. Quicksort has previously been considered an inefficient sorting solution for graphics processors, but we show that in CUDA, NVIDIA's programing platform for general-purpose computations on graphical processors, GPU-Quicksort 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.

A Practical Quicksort Algorithm for Graphics Processors

“…The source code of GPU-Quicksort is available for non-commercial use [21]. For benchmarking we used the following distributions which are defined and motivated in [22].…”

Performance improvement of data mining in Weka through multi-core and GPU acceleration: opportunities and pitfalls

“…For this reason, most implementations of GPU QuickSort like Cederman and Tsigas (2010) and Banerjee et al (2014) use a hybrid implementation where the CPU performs several scatter-gather calls to the GPU to perform the sort of a data array.…”

Parallel Olap on Multi/Many-Core and Cloud Platforms

“…We experimented with several bucket sizes and number of samples in order to best fit them to the GPU memory structure. For sorting the selected sample and the bottom level sorts of the individual buckets, we experimented with several existing GPU sorting methods such as bitonic sort, adaptive bitonic sort [58] based on [27], and parallel quick sort [30].…”

“…The thread block then sorts a sublist of n m data items in the SM's local shared memory. We tested different implementations for the local shared memory sort within an SM, including quick sort [30], bitonic sort, and adaptive bitonic sort [27]. In our experiments, bitonic sort was consistently the fastest method, despite the fact that it requires O(n log 2 n) work.…”

“…In Step 4, we sort all sm selected samples in global memory, using all available SMs in parallel. Here, we compared GPU bitonic sort [54], adaptive bitonic sort [58] based on [27], and gpu quick sort [30]. Our experiments indicate that for up to 16 million data items, simple bitonic sort is still faster than others due to its simplicity, small constants, and complete avoidance of conditional branching.…”