Interprocedural array redistribution data-flow analysis

Abstract

In High Performance Fortran (HPF), array redistribution can be described explicitly using directives (REDISTRIBUTE or REALIGN) which specify where new distributions become active or implicitly by calling functions which require different data distributions than the calling function. In order to actually compile an HPF program into an efficient form, however, both the redistribution operations as well as the possible distributions for the individual blocks of code must be known at compile-time. In this paper, we present an interprocedural data-flow framework which takes into account both explicit and implicit redistribution to automatically: (1) determine which distributions hold over specific sections of a program, (2) optimize both the inter- and intraprocedural transitions between dynamic distributions while still maintaining the original semantics of the HPF program, (3) determine when the distribution pattern specified by an HPF program causes a given array to be assigned multiple distributions due to different redistribution operations on multiple paths within a function or as a result of parameter aliasing (resulting in a non-conforming HPF program), as well as (4) convert (well behaved) dynamic HPF programs into equivalent static forms through a process we refer to as static distribution assignment (SDA) which can be used to extend the capabilities of existing subset HPF compilers that support static data distributions. As the approach presented in this paper has already been implemented as part of the PARADIGM (PARAllelizing compiler for DIstributed-memory General purpose Multicomputers) project at the University of Illinois, examples will also be presented to demonstrate several applications of this framework.

This research, performed at the University of Illinois, was supported in part by the National Aeronautics and Space Administration under Contract NASA NAG 1-613, in part by an Office of Naval Research Graduate Fellowship, and in part by the Advanced Research Projects Agency under contract DAA-H04-94-G-0273 administered by the Army Research office.