On the Implementation of Boundary Element Engineering Codes on the Cell Broadband Engine

  • Authors:
  • Manoel T. Cunha;Jose C. Telles;Alvaro L. Coutinho

  • Affiliations:
  • Civil Engineering Department / COPPE, Federal University of Rio de Janeiro, Rio de Janeiro, Brazil 21941-972;Civil Engineering Department / COPPE, Federal University of Rio de Janeiro, Rio de Janeiro, Brazil 21941-972;Civil Engineering Department / COPPE, Federal University of Rio de Janeiro, Rio de Janeiro, Brazil 21941-972

  • Venue:
  • High Performance Computing for Computational Science - VECPAR 2008
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

Originally developed by the consortium Sony-Toshiba-IBM for the Playstation 3 game console, the Cell Broadband Engine processor has been increasingly used in a much wider range of applications like HDTV sets and multimedia devices. Conforming the new Cell Broadband Engine Architecture that extends the PowerPC architecture, this processor can deliver high computational power embedding nine cores in a single chip: one general purpose PowerPC core and eight vector cores optimized for compute-intensive tasks. The processor's performance is enhanced by single-instruction-multiple-data (SIMD) instructions that allow to execute up to four floating-point operations in one clock cycle. This multi-level parallel environment is highly suited to applications processing data streams: encryption/decryption, multimedia, image and signal processing, among others. This paper discusses the use of Cell BE to solve engineering problems and the practical aspects of the implementation of numerical method codes in this new architecture. To demonstrate the Cell BE programming techniques and the efficient porting of existing scalar algorithms to run on a multi-level parallel processor, the authors present the techniques applied to a well-known program for the solution of two dimensional elastostatic problems with the Boundary Element Method. The programming guidelines provided here may also be extended to other numerical methods. Numerical experiments show the effectiveness of the proposed approach.