The selection of efficient implementations for a high-level language

This paper considers the problem of identifying an efficient set of implementations for the abstract constructs in a very high level program description. LIBRA is a system that prunes and expands a tree of partially implemented program descriptions, given a set of refinement rules for generating the tree. Several sets of rules group, order, and select refinements. The analysis of the cost of a program (or program part) at any level of refinement is maintained for cost comparisons between different refinements, for bottleneck identification, and for branch and bound search.