Toward Compiler Support for Scalable Parallelism Using Multipartitioning

Abstract

Strategies for partitioning an application’s data play a fun- damental role in determining the range of possible parallelizations that can be performed and ultimately their potential efficiency. This paper describes extensions to the Rice dHPF compiler for High Performance Fortran which enable it to support data distributions based on multi- partitioning. Using these distributions can help close the substantial gap between the efficiency and scalability of compiler-parallelized codes for multi-directional line sweep computations and their hand-coded coun- terparts. We describe our the design and implementation of compiler support for multipartitioning and show preliminary results for a bench- mark compiled using these techniques.

This work has been supported by NASA Grant NAG 2-1181, sponsored by DARPA and Rome Laboratory, Air Force Materiel Command, USAF, under agreement number F30602-96-1-0159, and supported in part by the Los Alamos National Laboratory Computer Science Institute (LACSI) through LANL contract number 03891-99-23, as part of the prime contract (W-7405-ENG-36) between the Department of Energy and the Regents of the University of California. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright annotation thereon. The views and conclusions contained herein are those of the authors and should not be interpreted as representing the official policies or endorsements, either expressed or implied, of DARPA and Rome Laboratory or the U.S. Government.