Lower Bound Cost Analysis

• Given mode and (upper approximation of) type information about procedures, the analysis automatically derives lower bounds on their computational costs (resolution steps).

• Estimation of lower bounds more complex than upper bounds (take into account possibility of failure).

• Failure of head unification leads to a trivial lower bound of 0.

• Improved non-failure analysis: given mode and (upper approximation of) type information, detects procedures (and calls) that can be guaranteed not to fail.

• Non-failure information is useful for:
  • Program transformations (execute possibly-failing goals ahead of non-failing goals).
  • Systems that exploit speculative parallelism.