Parallelization Based on Non-Strict Independence

Conclusions

• Non-strict independence increases the number of programs for which automatic exploitation of and-parallelism is feasible (and profitable)
• The evaluation shows that the new parallelizer completely subsumes the previous, strict independence based parallelizers:
  • When the program exhibits strict independence, it gives comparable results (intermediate between MEL and CDG)
  • When the program exhibits non-strict independence, it (obviously) shows improved results