HIGH PERFORMANCE FORTRAN FORUM ACHIEVES WIDE RECOGNITION AS IT ENTERS 1995

The High Performance Fortran Forum (HPFF) is a series of meetings lead by the CRPC since 1992 to define High Performance Fortran, an extension to Fortran 90 that provides support for portable data-parallel programming. Since its founding, much has been accomplished and commercial implementations of HPF version 1.0 are now appearing. Applied Parallel Research, Digital, Intel, Kuck and Associates, Meiko, Motorola, NEC, ACE, Hitachi, the Portland Group, Inc., and SofTech have already announced commercial products based on the HPF 1.0 standard (some of these are joint ventures). Archipel, Convex, Cray Computer, Fujitsu, IBM, Lahey, MasPar, NAG, NA Software, nCube, Pacific Sierra Research, and Thinking Machines have all announced commercial efforts to develop HPF-related products. In addition, there has been a great deal of academic work on data-parallel language systems.

The HPCC user community has been quite supportive of the HPF effort. A group of users and code developers interested in promoting convergence in the field of parallel computing sent an open letter to the parallel computing community stating that the data parallel paradigm, as embodied in High Performance Fortran (HPF), is capable of satisfying the needs of most supercomputing users. In the report they stated, "In our opinion, there should be convergence on HPF as the primary means for scientific parallel programming...What leads us to make such a bold statement? It is the conviction that only HPF will satisfy the three essential requirements for convergence: usability, applicability, and portability." The group encouraged vendors to view the HPF compiler as the primary vehicle for the development of parallel applications, asked supercomputer centers to provide HPF compilers on all of their parallel machines, and suggested that users move to HPF programming.

The annual HPFF meeting in January served as a forum to discuss recent developments in HPF, to propose new directions for the language, and encourage the use of the language. There are well-known limitations of the initial HPF definition, including no direct support for parallel I/O operations, limited support for irregular computations, and limited support for task parallelism. Many extensions to accommodate these limitations have been put forward, including some with practical implementation experience. This meeting provided an opportunity to present these proposals and discuss plans for continued work.

For more information, see the HPFF WWW page at http://www.crpc.rice.edu/HPFF/home.html.

Table of Contents