Source-Level Debugging of Automatically Parallelized Programs

Abstract

Parallelizing compilers automatically translate a sequential program into a parallel program. They simplify parallel programming by freeing the user from the need to consider the details of the parallel architecture and the parallel decomposition. A source-level debugger for automatically parallelized programs hides the parallelism from the user by providing the illusion that the original sequential program is executing. To provide a source-level view, a debugger has two tasks. The first task is to make it appear that programs execute operations in source order. The second task is to hide the decomposition of data and computation. This thesis describes a method for implementing dynamic order restoration and structural mapping in a debugger for automatically parallelized programs. Examples are taken from the domain of loop-based parallelism.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Oct 23, 1992
Accession Number
ADA261520

Entities

People

  • Robert S. Cohn

Organizations

  • Carnegie Mellon University

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Algorithms
  • Computations
  • Computer Programming
  • Computer Programs
  • Computer Science
  • Computers
  • Computing System Architectures
  • Debugging
  • High Level Languages
  • Language
  • Lisp Programming Language
  • Machine Languages
  • Operating Systems
  • Parallel Computing
  • Parallel Processing
  • Programming Languages
  • Two Dimensional

Fields of Study

  • Computer science
  • Engineering

Readers

  • Parallel and Distributed Computing.