DETECTION OF IMPLICIT COMPUTATIONAL PARALLELISM FROM INPUT-OUTPUT SETS.
Quarterly technical rept. no. 1, 15 Jul-15 Oct 66,
BURROUGHS CORP PAOLI PA DEFENSE SPACE AND SPECIAL SYSTEMS GROUP
Pagination or Media Count:
This is the first report of an investigation designed to show how the implicit parallelism in programs written in present programming languages can be recognized and exploited by computers with highly parallel machine organizations. A language-independent recognition algorithm is developed. A program is treated as a partially ordered set of processes. A process is a transformation of an input set into an output set. The algorithm represents a method for converting the given ordering relation among processes in any given program into the essential ordering relation. Unnecessary serial ordering imposed by present programming languages is eliminated. The essential ordering is recognized by comparing inputs to processes with outputs of selected prior processes. The number of comparisons is the minimum necessary to detect the essential ordering. The algorithm is explained in detail and the equivalent graphical operations are described. Program loops and conditionals can be analyzed within the framework of the algorithm. Process inputs and outputs are related to memory and input-output devices. Various levels of partitioning a program into processes are discussed. Author
- Computer Programming and Software
- Computer Hardware