The Portland Group Inc, Wilsonville, Oregon, and Syracuse University’s Northeast Parallel Architectures Center have formed a technology alliance to research, develop and distribute compilers and tools for parallel computer systems. First fruit of the agreement, announced this week’s in Minneapolis, Minnesota, is a high-performance Fortran compiler targeted at Multiple Instruction-Multiple Data and Single Instruction Multiple Data systems. The High Performance Fortran compiler includes a set of extensions to the Fortran 90 language, allowing programmers to create machine-independent software optimised for a range of parallel architectures. Targets set for the compiler – including provision of a single parallel programming model for developing applications that can run on a variety of computers – were formulated by a coalition of industrial groups known as the High Performance Fortran Forum. The High Performance Fortran compiler, demonstrated at the supercomputing show, is based on a data parallel language – an optimising Fortran90D compiler – developed by Syracuse group and backed by funding from the US Defense Advanced Research Projects Agency. The initiative is part of a wider effort to facilitate the development of applications that will run on future generations of parallel architectures without loss of performance. Maryland and Rice Universities are also contributing to the work, which is being carried out as part of the National Science Foundation’s Center for Research in Parallel Computation. The Syracuse group is developing prototypes of software technologies which will accelerate the take-up of general purpose parallel computing. Its alliance with the Portland Group will result in future data parallel compilers for other languages such as C++ and Ada, as well as communications interfaces and run-time tools, which will be targeted at networks and clusters of heterogeneous machines and shared-memory parallel computers. The High Performance Fortran compiler includes parallel loop analysis, inter-procedural analysis, inter-node communication optimisations and vectorisation of Fortran 90 style array constructs and Fortran 77 indexed array constructs. High Performance Fortran is input to the compiler, which produces standard Fortran 77 and message-passing calls as output. Output from the compiler is compiled by standard, single-node Fortran 77 compilers. According to the Portland Group the High Performance compiler can be targeted for sue by all sorts of parallel systems and clusters of workstations on a network. The compiler will be out late next year.
