HPF: Achievements, Problems, and Prospects

HPF has reached a critical stage in its history. Having struggled while the compiler technology evolved into a usable state, the community has now found it possible to write portable, high-performance implementations for selected applications in HPF. However, there are a number of pitfalls ahead. Notable among these are the general drift away from Fortran for advanced applications, remaining difficulties with HPF optimization technology, inadequacy of support for irregular problems in current implementations, the continuing need for advanced mathematical library support, and the emergence of competing paradigms such as MPI and OpenMP. The HPF user and vendor community must deal with these issues in the near future if the language is to emerge as the option of choice for high-performance scientific computing on scalable parallel machines. The talk will conclude with some strategies for addressing these issues.