Integrating parallel dataflow programming with the Ada tasking model

  • Authors:

  • John Thornley

  • Affiliations:

  • Computer Science Department, California Institute of Technology, California

  • Venue:
  • TRI-Ada '94 Proceedings of the conference on TRI-Ada '94
  • Year:
  • 1994

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper describes how parallel dataflow programming can be simply and efficiently integrated with the Ada tasking model. Three extensions to standard Ada are proposed: parallel composition of statements, a parallel for-loop statement, and single-assignment types. Using these constructs, parallel Ada programs can be written without explicit tasking, that are identical—except for two new reserved words—to sequential Ada programs that satisfy the same specifications. These programs can be developed as sequential programs, then executed as deterministic parallel programs for high performance. The proposed extensions are defined by transformation into standard Ada tasking, and can efficiently be implemented by straightforward preprocessing, without any change to the underlying Ada compiler or run-time system.