Performance Issues of SYRK Implementations in Shared Memory Environments for Edge Cases

Md Mosharaf Hossain, Thomas M. Hines, Sheikh K. Ghafoor, Ryan J. Marshall, Muzakhir S. Amanzholov, Ramakrishnan Kannan

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

Abstract

The symmetric rank-k update (SYRK) is a level-3 BLAS routine commonly used by many Data Mining/Machine Learning(DM/ML) algorithms such as regression, dimensionality reduction algorithms like PCA, matrix factorization and k-mean clustering. This paper presents a comprehensive analysis of the SYRK routine under popular dense linear algebra libraries such as OpenBLAS, Intel MKL, and BLIS particularly focusing on edge cases of dense matrices (thin or fat shapes) that are common in DM/ML applications. Our work identifies some performance issues of the SYRK routine in multi-threaded shared memory environments for edge cases and discuss matrix dependent modifications for performance improvement.

Original languageEnglish
Title of host publication2018 21st International Conference of Computer and Information Technology, ICCIT 2018
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781538692424
DOIs
StatePublished - Jul 2 2018
Event21st International Conference of Computer and Information Technology, ICCIT 2018 - Dhaka, Bangladesh
Duration: Dec 21 2018Dec 23 2018

Publication series

Name2018 21st International Conference of Computer and Information Technology, ICCIT 2018

Conference

Conference21st International Conference of Computer and Information Technology, ICCIT 2018
Country/TerritoryBangladesh
CityDhaka
Period12/21/1812/23/18

Funding

This manuscript has been authored by UT-Battelle, LLC under Contract No. DE-AC05-00OR22725 with the U.S. Department of Energy. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or reproduce the published form of this manuscript, or allow others to do so, for United States Government purposes. The Department of Energy will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public-access-plan).

Keywords

  • BLAS
  • Multicore
  • Performance Issues of SYRK
  • Scalability of AA

Fingerprint

Dive into the research topics of 'Performance Issues of SYRK Implementations in Shared Memory Environments for Edge Cases'. Together they form a unique fingerprint.

Cite this