EFFICIENT IMPLEMENTATION OF A HIGH-LEVEL LANGUAGE ON A BIT-SERIAL PARALLEL MATRIX PROCESSOR
Abstract
Many modern supercomputers perform operations in parallel (more than one at once) to increase processing power. On important type of parallel processor is the bit-serial parallel matrix processor, an example of which is the NASA's Massively Parallel Processor. Programming a parallel matrix processor requires languages which are capable of directly expressing those operations which the machine can support. A good language should be powerful, easy to use, and efficiently implementable. This thesis describes the development of the language Parallel Pascal for the class of parallel matrix processors, and describes the implementation of Parallel Pascal by a two-phase compiler whose phases communicate through an intermediate language called Parallel P-code. The efficient implementation of Parallel Pascal by this means is examined relative to the Massively Parallel Processor. System performance as a function of input-output, and the construction of efficient specialized bit-serial functions, are also examined. Parallel Pascal is shown to be well-suited to the MPP, and the use of Parallel P-code as an intermediate language is shown to be a workable implementation. The implementation of a simple input-output system is described. A heuristic method for generating specialized bit-serial functions is presented.
Degree
Ph.D.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.