Int. J. Computational Science and Engineering, Vol. 20, No. 4, 2019 501
Copyright © 2019 Inderscience Enterprises Ltd.
CUDA GPU libraries and novel sparse matrix-vector
multiplication – implementation and performance
enhancement in unstructured finite element
computations
Richard Haney
The MITRE Corporation,
7515 Colshire Drive,
McLean, Virginia, 22102-7539, USA
Email: rhaney@mitre.org
Ram Mohan*
North Carolina A&T State University,
2907 East Gate City Blvd.,
Greensboro, North Carolina, 27401, USA
Email: rvmohan@ncat.edu
*Corresponding author
Abstract: The efficient solution to systems of linear and nonlinear equations arising from sparse
matrix operations is a ubiquitous challenge for computing applications that can be exacerbated by
the employment of heterogeneous architectures such as CPU-GPU computing systems. This
paper presents our implementation of a novel sparse matrix-vector multiplication (a significant
compute load operation in the iterative solution via pre-conditioned conjugate gradient based
methods) employing LightSpMV with compressed sparse row (CSR) format, and the resulting
performance characteristics using an unstructured finite element-based computational simulation.
Computational performance analysed indicates that LightSpMV can provide an asset to boost
performance for these computational modelling applications. This work also investigates
potential improvements in the LightSpMV algorithm using CUDA 35 intrinsic, which results in
an additional performance boost by 1%. While this may not be significant, it supports the idea
that LightSpMV can potentially be used for other full-solution finite element-based computational
implementations.
Keywords: general purpose GPU computing; GPGPU; sparse matrix-vector; finite element
method; FEM; Compute Unified Device Architecture; CUDA; performance analysis.
Reference to this paper should be made as follows: Haney, R. and Mohan, R. (2019)
‘CUDA GPU libraries and novel sparse matrix-vector multiplication – implementation and
performance enhancement in unstructured finite element computations’, Int. J. Computational
Science and Engineering, Vol. 20, No. 4, pp.501–507.
Biographical notes: Richard Haney received his BS in Computer Science and PhD in
Computational Science and Engineering with a focus on GPGPU and High Performance
Computing in 2013. Currently, he is a Senior Computer Scientist at The MITRE Corporation
J84B Simulation Engineering division. His research interests include distributed high
performance computing and GPGPU.
Ram Mohan is currently a Professor of Nanoengineering with Department of Nanoengineering in
Joint School of Nanoscience and Nanoengineering at North Carolina A&T State University. He
leads the computational nanoengineering focus area with research interests in physics-based
modelling and enabling scalable, high performance computational modelling developments,
performance, and their applications.
1 Introduction
As heterogeneous architectures become the standard
paradigm for high performance computing (HPC), interest
in efficient sparse matrix-vector multiplication using
graphics processing units (GPU) has increased. The sparse
matrix-vector multiplication is the largest cost of many
physics-based modelling solutions that utilise finite element
(FE) or finite difference (FD) discretisation. These methods
result in a need to solve a system of linear equations using
iterative methods such as conjugate gradient (CG) or other