Parallelization Based on Non-Strict Independence

• Need of a global analysis of the program (e.g. by Abstract Interpretation): conditions not wholly checkable at run-time

• But run-time tests can complete this information:
  • allvars(X,): every free variable in X is in the list (ground(X) = allvars(X,[]))
  • sharedvars(X,Y,): every free variable shared by X and Y is in the list (indep(X,Y) = sharedvars(X,Y,[]))

• Compile-time required information: run-time instantiation of the variables, at least freeness, sharing and groundness
  • Starting point: Sharing+Freeness analysis
  • Techniques easily applied to more powerful analysis
    (e.g. Depth-K Sharing).

• Important point: the techniques developed for NSI subsume the previously proposed for SI - and do not need new machinery to run.