Fast radix sort for sparse linear algebra on GPU

Lukas Polok, Viorela Ila, Pavel Smrz

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Fast sorting is an important step in many parallel algorithms, which require data ranking, ordering or partitioning. Parallel sorting is a widely researched subject, and many algorithms were developed in the past. In this paper, the focus is on implementing highly efficient sorting routines for the sparse linear algebra operations, such as parallel sparse matrix - matrix multiplication, or factorization. We propose a fast and simple to implement variant of parallel radix sort algorithm, suitable for GPU architecture. Extensive testing on both synthetic and real-world data shows, that our method outperforms other similar state-of-the-art implementations. Our implementation is bandwidth-efficient, as it achieves sorting rates comparable to the theoretical upper bound of memory bandwidth. We also present several interesting code optimizations relevant to GPU programming.

Original languageEnglish
Title of host publicationProceedings of the 2014 Spring Simulation Multiconference, SpringSim 2014 - High Performance Computing Symposium, HPC 2014
PublisherThe Society for Modeling and Simulation International
Pages79-86
Number of pages8
Edition5
ISBN (Print)9781632662163
Publication statusPublished - 2014
Externally publishedYes
Event22nd High Performance Computing Symposium, HPC 2014, Part of the 2014 Summer Simulation Multiconference, SummerSim 2014 - Tampa, FL, United States
Duration: 13 Apr 201416 Apr 2014

Publication series

NameSimulation Series
Number5
Volume46
ISSN (Print)0735-9276

Conference

Conference22nd High Performance Computing Symposium, HPC 2014, Part of the 2014 Summer Simulation Multiconference, SummerSim 2014
Country/TerritoryUnited States
CityTampa, FL
Period13/04/1416/04/14

Fingerprint

Dive into the research topics of 'Fast radix sort for sparse linear algebra on GPU'. Together they form a unique fingerprint.

Cite this