An Application Programming Methodology for Multiprocessors

The programming of applications for multiprocessors poses a considerable challenge to application programmers incurring penalties on their time. These difficulties are compounded by the need to redesign the algorithms and rewrite the applications when presented with a different architecture. One possible method to ease this difficulty is to express new applications in terms of a number of smaller parallel algorithms which have been previously implemented efficiently. This talk describes two software systems - CAPE and PARAL currently underway, which help alleviate this problem. PARAL is intended to be a collection of basic parallel algorithms. CAPE provides the programmer with a set of programming utilities useful for combining parallel algorithms. Some preliminary results from an initial implementation of the system on NCUBE are also described.