Bank Conflict Free Comparison-based Sorting On GPUs

TL;DR

This paper introduces a novel GPU sorting framework that eliminates memory bank conflicts in shared memory, enabling more efficient comparison-based sorting and merging of large data streams.

Contribution

It presents the first comparison-based shared memory sorting algorithm on GPUs that is free of bank conflicts, improving performance for GPU sorting tasks.

Findings

Developed a bank conflict free shared memory sorting algorithm.

Designed BCFMergesort for merging large sorted streams without conflicts.

Achieved coalesced global memory accesses during merging.

Abstract

In this paper we present a framework for designing algorithms in shared memory of GPUs without incurring memory bank conflicts. Using our framework we develop the first comparison-based shared memory sorting algorithm that incurs no bank conflicts. It can be used as a subroutine for GPU sorting algorithms to replace current use of sorting networks in shared memory. Using our bank conflict free shared memory sorting subroutine as a black box, we design BCFMergesort, an algorithm for merging sorted streams of data that are larger than shared memory. Our algorithm performs all accesses to global memory in coalesced manner and incurs no bank conflicts during the merge.