Optimizations based on Analysis Information (I)

• Program simplification
  • Analysis information is used in automatic parallelization to
    • guide the annotation process (heuristics)
    • eliminate run-time tests
    in clauses that are parallelized
  • Analysis information can be used to
    • eliminate dead-code
    • eliminate redundant tests (sure success)
    • reduce tests to fail (sure failure)
    at any program point
  • Parametric with respect to abstract domain
  • Table of ``Abstract Executability''

• Multiple specialization
  • Generating several versions of a predicate for different uses
  • Allows further optimizations
  • Problems and solutions
    • Creating versions multivariant analysis.
    • Increase in code size minimizing.
    • Selecting appropriate version renaming (no run-time checks needed).