Coupling Parallel Data and Work Partitioners to VFCS

A significant amount of software research for developing programming environments for distributed--memory systems (DMSs) is currently underway in both academia as well as industry. This paper addresses the automatic parallelization of irregular codes for DMSs. We present the language constructs provided by Vienna Fortran to deal efficiently with irregular codes. They have been implemented within the Vienna Fortran Compilation System. This system also provides an interactive interface which enables the user to direct the system to derive data and work distributions automatically according to a selected strategy. The description of these advanced features is the main concern of the paper. 1 Introduction Languages like Vienna Fortran (VF) [8], Fortran D/90D [4], or High Performance Fortran (HPF) [5], that are based on the data parallel programming model, represent a modern approach to programming DMSs. Here, programs are written using constructs of a sequential language to specify the com..