Effective Optimization of Multiple Traversals in Lazy Languages

Tupling transformation strategy can be applied to eliminate redundant calls in a program and also to eliminate multiple traversals of data structures. While the former application can produce super-linear speedup in the transformed program, the eeectiveness of the latter has yet to be appreciated. In this paper, we investigate the pragmatic issues behind elimination of multiple data traversal in the context of lazy languages, and propose a framework of tupling tactic called strictness-guided tu-pling. This tactic is capable of exploiting specialised strictness contexts where possible to eeect tupling op-timisation. Two further enhancements of the tupling tactic are also proposed. One achieves circular tupling when multiple traversals from nested recursive calls are eliminated. The other exploits speculative strictness to further improve the performance of tupling. Benchmarks are given throughout the paper to illustrate the performance gains due to these tactics.