3.B.4 SimpleTechniques for Debugging

• You have a parallel run with wrong answers

• Is sequential version different?

  • Usually sequential version is considered right
  • Is it?

• Check parallel P=1 results

  • This isolates problems like data races

• Vary the scheduling

• Binary chop large programs to isolate

  • Breakdown large parallel regions
  • Compile with directives off in half

Previous slide

Next slide

Back to first slide

View graphic version