Efficient Execution of Declarative Programs

Memoization is an optimization that provides asymptotic speedups, automatically achieving many of the benefits of dynamic programming. Memoization, however, trades off reduced execution time for additional required storage. This additional storage requirement can be reduced somewhat by using several techniques from incrementalization. The first technique, called pruning, statically identifies the set of cached results which are still required at any particular program point, allowing the remainder to be discarded. The second technique dynamically tracks dependencies, allowing a cached result to be reused in a broader set of contexts. This paper suggests that it would be possible to provide a generic graph library with a declarative interface in a traditional imperative programming language such as Java or C++. This library could be implemented using memoization and a dynamic version of pruning. This would enable users of traditional programming languages to program at a higher level of abstraction while still achieving the efficiency they require.