PLAN FOR DETECTION OF PARALLELISM IN COMPUTER PROGRAMS.
Abstract
This is the third report of an investigation to determine how implicit parallelism in programs written in compiler languages can be recognized and exploited by machines with highly parallel organizations. The algorithm developed in the two prior reports for detecting the essential order among program parts based upon input-output set intersections and any initially known essential ordering has been applied in this report to loop and array structures. Although parallelism in many array references can be detected, run-time index determination limits the general recognition of array element parallelism. The effects of data structures on parallelism are related to machine organization and memory accessing. The block diagram of a parallelism detection program is given and the source language for programs to be analyzed is described. This program will be used as a tool for the demonstration of the desirability and feasibility for automatic recognition of parallelism in compiler language programs. Levels of program partitioning are discussed. A method for identifying the different instances of a variable during a program and their candidacy as inputs for subsequent references is described. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1967
- Accession Number
- AD0655867
Entities
People
- David A. Fisher
- Harvey W. Bingham
- John W. Seward