Flow-Directed Lightweight Closure Conversion

This paper presents a lightweight closure-conversion method that is driven by the results of wholeprogram interprocedural flow, reachability, points-to, and escape analyses. The method has been implemented and evaluated as part of a complete Scheme compiler. When compared with a baseline closure-conversion method that does no optimization, as well as conventional closureconversion methods that only do optimizations that do not rely on interprocedural analysis, this method significantly increases the degree of closure, variable-slot, parent-slot, closure-pointer– slot, variable, parent-parameter, parent-passing, closure-pointer, variable-spilling, and parentspilling elimination. It also significantly increases the degree of parent-slot, closure-pointer–slot, and parent-parameter compression and reduces the number of indirections per variable reference. Finally, code produced by this compiler runs significantly faster than code produced by other state-of-the-art Scheme compilers.