Skip to main navigation Skip to search Skip to main content

An improved sparse matrix-vector multiplication kernel for solving modified equation in large scale power flow calculation on CUDA

  • Mei Yang*
  • , Cheng Sun
  • , Zhimin Li
  • , Dayong Cao
  • *Corresponding author for this work

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

Abstract

Sparse matrix-vector multiplication (SpMV) is the most important kernel in parallel iterative method for solving modified equation in large scale power system power flow calculation. In this paper, one improved compressed sparse row (ICSR) storage used to settle the problem of the global memory alignment in the vector kernel on Graphics processing Unit (GPU) is given. The experiments on matrices with different sizes demonstrate that the vector kernel with ICSR storage format could improve the performance by 5%-30% for SpMV comparing with vector kernel with CSR, especially for the large-scale unstructured sparse matrix-vector product, the effect is more obvious.

Original languageEnglish
Title of host publicationConference Proceedings - 2012 IEEE 7th International Power Electronics and Motion Control Conference - ECCE Asia, IPEMC 2012
Pages2028-2031
Number of pages4
DOIs
StatePublished - 2012
Event2012 IEEE 7th International Power Electronics and Motion Control Conference - ECCE Asia, IPEMC 2012 - Harbin, China
Duration: 2 Jun 20125 Jun 2012

Publication series

NameConference Proceedings - 2012 IEEE 7th International Power Electronics and Motion Control Conference - ECCE Asia, IPEMC 2012
Volume3

Conference

Conference2012 IEEE 7th International Power Electronics and Motion Control Conference - ECCE Asia, IPEMC 2012
Country/TerritoryChina
CityHarbin
Period2/06/125/06/12

Keywords

  • CUDA
  • Compressed sparse row (CSR) storage format
  • GPU
  • Parallel algorithm
  • Sparse matrix-vector multiplication
  • modified equation
  • power flow calculation

Fingerprint

Dive into the research topics of 'An improved sparse matrix-vector multiplication kernel for solving modified equation in large scale power flow calculation on CUDA'. Together they form a unique fingerprint.

Cite this