We have extended an existing HPF compiler with language features designed for the parallelization of unstructured computations on multicomputers.The language extensions include block-general distributions and dynamic data distributions specified through userdefined mapping arrays and finctions.A prototype compiler has been implemented which features dlfherent run-time preprocessing mechanisms and also allows clean integration of explicit messagepassing primitives.The compiler is developed as part of a complete multicomputer programming environment being used by a group of application developers in the framework of the Joint CSCS-ETH/NEC Collaboration in Parallel Processing.As such, it is supported by a high-level debugger andpe~oimance monitoc and the usability and efficiency of generated parallel programs is validated by the application developers.In this pape~we summarize the programming paradigm implemented through HPF extensions, and detail the respective compiler directives.We describe the implemented run-time preprocessing mechanisms and evaluate the efjlciency of compiler-generated code on an NEC Cenju-3 multicomputez

Distributed memory parallel processors (DMPPs) have no hardware support for a global address space. However, conventional programs written in a sequential imperative language such as Fortran typically manipulate few, large arrays. The Oxygen compiler, developed as part of the K2 project, translates conventional Fortran code, augmented with code and data distribution directives, into C programs including SEND/RECEIVE communication primitives. The compiler directives, which are either supplied by the user, or for simple programs generated automatically, support a global name space through a run-time mechanism called data consistency analysis. We report in this paper the performance of seven parallel programs generated by Oxygen for three DMPPs, namely for a Parsytec Supercluster, an iWARP, and for the Fujitsu AP1000. All machines were configured as 8 x 8 tori.

This paper presents some preliminary results toward the automatic parallelization of uniprocessor FORTRAN code on distributed-memory parallel processors (DMPPs). The paper introduces Oxygen, a compiler for a DMPP under development at the Laboratory. The design of Oxygen and its parallelization strategy are discussed, and an analysis of its most significant components is presented, together with performance benchmarks. Oxygen carries out data consistency analysis at run-time; our results show that the overhead introduced is acceptable. Run-time data consistency analysis may also be the only viable approach to parallelize certain “hard” algorithms, as we will show in this study.

We have simulated several numerical and non-numerical algorithms on five distributed-memory parallel processors (DMPPs). All five DMPPs have the same topology (a torus), and the same number of nodes. The architectures differ only in the communication speed between neighboring nodes, while the computation unit is kept unchanged. The goal of the paper is to quantify the effect that interprocessor communication speed and synchronization overhead have on the performance of the DMPPs. After introducing the rationale for this study and reviewing related work, we present and discuss the results of the simulations.

The multiprocessor Sequent Symmetry was first delivered to customers with write-through caches. Later on each machine was upgraded with copy-back caches. Because all the other architectural parameters were unchanged (main memory, bus, cache organization and size, and so on), this made it possible to measure the performance of a multiprocessor with no caches, write-through caches, and copy-back caches. We also study the impact that the language (FORTRAN and C) has on the performance of the machine.