Home // COMPUTATION TOOLS 2015, The Sixth International Conference on Computational Logics, Algebras, Programming, Tools, and Benchmarking // View article


Combining Code Refactoring and Auto-Tuning to Improve Performance Portability of High-Performance Computing Applications

Authors:
Chunyan Wang
Shoichi Hirasawa
Hiroyuki Takizawa
Hiroaki Kobayashi

Keywords: auto-tuning; code refactoring; high-performance computing; performance portability

Abstract:
An existing High-Performance Computing (HPC) application is usually optimized for a particular platform to achieve high performance. Hence, such an application is often unable to run efficiently on other platforms, i.e., its performance is not portable. The purpose of this work is to establish a systematic way to improve performance portability of HPC applications, to which various kinds of platform-specific optimizations have already been applied. To this end, we combine code refactoring and auto-tuning technologies, and develop a programming tool for HPC refactoring. Auto-tuning is a promising technology to enable an HPC application to adapt to different platforms. In general, however, an auto-tuning tool is not applicable to an HPC application if the application has already been optimized for a particular platform. In this work, a code refactoring tool that interactively asks a user for necessary information to undo some platform-specific optimizations in an existing application is developed based on Eclipse, and hence auto-tuning techniques can be applied to the application. The evaluation results demonstrate that combining code refactoring and auto-tuning is a promising way to replace platform-specific optimizations with auto-tuning annotations, and thereby to improve the performance portability of an HPC application.

Pages: 20 to 26

Copyright: Copyright (c) IARIA, 2015

Publication date: March 22, 2015

Published in: conference

ISSN: 2308-4170

ISBN: 978-1-61208-394-0

Location: Nice, France

Dates: from March 22, 2015 to March 27, 2015