Moving scientific codes to multicore microprocessor CPUs

Paul R Woodward, Jagan Jayaraj, Pei Hung Lin, Pen-Chung Yew

Research output: Contribution to journalArticlepeer-review

20 Scopus citations

Abstract

Several significant that need to be followed for code transformation techniques, using the ;Piecewise-Parabolic Method (PPM) gas dynamic algorithm for scientific computing are discussed. The implementation of the PPM gas dynamics algorithm for code transformation techniques can improve the performance of scientific simulation codes in the latest IBM Cell processor and other multicore CPUs. The Cell is used for implementing the algorithm in scientific computing, as it has eight cores, along with an additional core to manage these cores. The features left out of the Cell SPE and the small size of its private cable enable more of these cores to be accommodated on a single chip that will be possible for general-purpose CPUs. This enables programmers, to handle the extra difficulty and receive the benefit, when a scientific application can be operated without the need for features omitted from Cell SPE.

Original languageEnglish (US)
Article number4653201
Pages (from-to)16-25
Number of pages10
JournalComputing in Science and Engineering
Volume10
Issue number6
DOIs
StatePublished - Nov 2008

Bibliographical note

Funding Information:
This work was supported by US Department of Energy (DoE) Office of Science grant DE-FG0203ER25569 from the Mathematical, Informational, and Computational Sciences (MICS) program and by a contract from Los Alamos National Laboratory through the DoE Advanced Simulation & Computing (ASC) program. We appreciate discussions during this work with Peter Hofstee, Karl-Heinz Winkler, David Porter, Robert Lowrie, and Ben Bergen. We thank Ben Bergen and Stephen Hodson for helping us run our test programs on many occasions on preproduction Cell hardware at Los Alamos during Fall 2006. We also acknowledge the donation to our lab, the LCSE, of two dual-Cell blades from IBM in January 2007 and the support for continued work with multicore CPUs from US National Science Foundation grant CNS-0708822. We also gratefully acknowledge support from the Minnesota Supercomputer Institute and, earlier, from the Pittsburgh Supercomputer Center for our work on interactive supercomputing.

Copyright:
Copyright 2011 Elsevier B.V., All rights reserved.

Fingerprint

Dive into the research topics of 'Moving scientific codes to multicore microprocessor CPUs'. Together they form a unique fingerprint.

Cite this