NPAC Technical Report SCCS-327

Applications Benchmarking Set for Fortran-D and High Performance Fortran

A Mohmad, Geoffrey Fox, Gregor vonLaszewski, Manish Parashar, Tomasz Haupt, Kim Mills, Ying-Hua Lu, Neng-Tan Lin, Nang-kang Yeh

Submitted June 26 1992


Abstract

Fortran77, the currently accepted Fortran standard worldwide, is essentially a sequential language which hides the parallelism of a problem in sequential constructs like loops, etc. Consequently, scientists wishing to use parallel computers must rewrite their programs in an extension of Fortran that explicitly reflects the architecture of the underlying machine, such as a message-passing dialect for MIMD distributed-memory machines, array syntax for SIMD machines, or an explicitly parallel dialect with synchronization for MIMD shared-memory machines. This conversion is difficult, error prone, and the resulting parallel codes are machine-specific. To overcome these problems a new Fortran standard, or more precisely, a standard of Fortran extensions, are necessary to establish a machine independent programming model that is easy to use and yet, is acceptably efficient on different parallel architectures. Thus research is now concentrated on the provision of appropriate high-level language constructs to enable users to design programs in much the same way as they are accustomed to on a sequential machine. Several proposals have been put forth in recent months for a set of language extension to achieve this. To coordinate these efforts the High Performance Fortran Forum (HPFF) has been created. HPFF is a coalition of industrial and academic groups working to develop an industry-wide standard of extensions to Fortran which provide support for high performance programming on a wide variety of machines, portable from workstations to massively parallel SIMD and MIMD supercomputers. Fortran-D, a version of Fortran enhanced with data decomposition directives, can provide such a programming model. We believe that it can make parallel computing truly usable. One of the elements of this program is to establish a reliable validation strategy. In order to evaluate the efficiency of automatic data partitioning schemes, a dedicated benchmarking suite is being developed at NPAC and is described in this paper. Currently, the suite is oriented towards validation of the Fortran-D compiler, which is also being developed at NPAC in collaboration with Rice University. We plan to augment it with applications written in other proposed HPF dialects.


PostScript version of the paper