Language Libraries and Performance Evaluation Tools for Scalable Parallel Systems

摘要

The primary objective of this project has been to develop a new programming language, compiler and programming environment, called pC++, that is based on a simple extension to C++ to support the development of software for Massively Parallel Processing (MPP) computer systems. This tool supports a platform for parallel object oriented software capable of running without modification on all commercial Multiple Instruction Multiple Data (MIMD) systems; (1) an interface to Single Program Multiple Data (SPMD) libraries such as ScaLapack++, A++ and POOMA; (2) an interface to High Performance Fortran (HPF); (3) an interface to control parallel C++ based languages such as CC++; (4) a way to exploit parallel I/O systems and persistent object databases; and (5) a complete programming environment. pC++ has been implemented on a wide range of commercially available parallel systems including the Thinking Machines CM-5, the Intel Paragon, the SGI PowerChallenge, the IBM SP-2 and the Cray T3-D. One of the most interesting by products of the pC++ project has been a language preprocessor toolkit called Sage++. This toolkit has been extended in a variety of ways and is used for a large number of applications, including the TAU programming environment. There have been four major components to C++ technology. The first component is the pC++ language. The mot important of these are applications that are part of the NSF NCSA Alliance. The second component of technology is the Sage++ compiler toolkit that was built for pC++. The third component of technology transfer is the Tulip runtime system, the ASCI Blue Mountain system. The fourth component of the technology is the TAU programming environment tools and will be the standard foundation for HPC++ tools.