Reports

The Semi-Automatic Generation of Processing Element Control Paths for Highly Parallel Machines

Sabety, Theodore M.; Mathies, Brian; Shaw , David Elliot

This paper describes a recently implemented program that very rapidly generates control paths for different variants of the constituent processing elements of a particular massively parallel machine, the NON-VON Supercomputer. The program, called PLATO, accepts as input a set of instruction opcodes, together with associated control information, and produces as output a functionally correct, highly area-efficient set of PLA's for the processing elements. One novel aspect of the program is its use of a channel routing algorithm to generate a Weinberger Array layout for the OR-plane of the PLA. By supporting extremely rapid generation of processing elements with different instruction sets, PLATO facilitates "rapid turnaround" architectural experimentation of a sort that would otherwise be impractical. Use of the program has already yielded major area and performance improvements in the NON-VON processing element. Many of the techniques employed in the PLATO system should prove applicable to the semi-automatic layout of processing elements for other multiprocessor machines.

Subjects

Files

More About This Work

Academic Units
Computer Science
Publisher
Department of Computer Science, Columbia University
Series
Columbia University Computer Science Technical Reports, CUCS-127-84
Published Here
February 17, 2012