Abstract

This paper describes the results of the research for implementing applications for concurrent execution on the CYBERPLUS multiparallel computer system. Three layers of parallelism are built into this system: instruction, computation and functional levels. The paper contains a short summary of the hardware, and the different layers of the software for multiprocessor systems. They are based on the previously developed CPFTN (Cyberplus Fortran Compiler) compiler which produces optimized object code for a single CYBERPLUS processor, exploiting through a series of horizontal and vertical optimizations the internal parallelism present in this processor. Unlike other compilers, the input to CPFTN consists of an entire kernel containing all subroutines and functions loaded into a single processor. An interprocessor communications subsystem provides the low-level component of the multiprocessor system and contains the data transfer and synchronization capabilities. The high-level component consists of extensions to FORTRAN (CPMFTN, Cyberplus Multiprocessor Fortran) for characterizing the data shared between processors and implements concepts specially adapted to the private memory architecture of the CYBERPLUS as well as to the specific needs of the user community targetted by this product. The resulting scheme is a large grain size, demand driven, data flow type. It is assumed that partition between processors can best be done on the bases of the high-level knowledge of the problem possessed by the user. The corresponding kernels may be created independently of each other, using the new language features introduced to characterize data shared between processors. Scheduling, synchronization, as well as data transfer operations are automatically distributed by the compiler across the subroutines and functions constituting the grain size of the partition. The first version of CPMFTN under development is described followed by several simple application paradigms and discussion of the merits of possible future extensions.

Full Text
Paper version not known

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call

Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.