An HPF Case Study of a Domain-Decomposition Based Irregular Application

. Data-parallel languages, in particular HPF, provide a highlevel view of operators overs parallel data structures and hide the details of data partitioning and communication. One of the most difficult issues in compiling such languages is managing irregular data-dependent parallelism. This paper presents the study of a realistic, but non adaptive irregular application. We show that HPF can easily express the natural parallelism of the application. Experimental results and a detailed examination of the compiler process are presented. 1 Introduction Data parallel languages, and especially High Performance Fortran (HPF) [1, 2] have been designed to make possible efficient and high-level parallel programming for distributed memory machines, in contrast to the lowlevel concurrent programming model based on explicit message-passing primitives. Even if many academic or proprietary data-parallel languages preceded it, HPF is very recent: the first language specification was completed in 93, ..