CMSC 631 Composing Dataflow Analyses and Transformations - Report

This paper presents a framework that allows several independent dataflow analyses which perform optimizations to be composed and applied concurrently. This is done so that every analysis will benefit from the transformations made to the program by all other analyses. Dataflow analyses are mostly used in optimizing compilers, to deduce transformations that can be applied on the original program, for optimization purposes, without changing the program’s behavior. Obviously, different analyses can benefit from each other’s results or transformations. Therefore, we want the analyses to be able to “communicate,” in order to maximize the optimizations applied to the program. Instead of manually writting a “combined” analysis by hand, in [3] a framework is proposed that tries to automatically combine various analyses. That way, each analysis is independently implemented, and it’s the framework’s responsibility to combine the analyses in a correct way.