Dataflow Optimization Made Simple

We present a Haskell library that makes it easy for compiler writers to implement program transformations based on dataflow analyses. The compiler writer must identify (a) a family of logical assertions on which the transformation will be based; (b) an approximate representation of such assertions, which must have a lattice structure such that every assertion can be increased at most finitely many times; (c) transfer functions that approximate weakest preconditions or strongest postconditions over the assertions; and (d) rewrite functions whose soundness is justified by the assertions. To guide compiler writers, we show how dataflow analyses are related to seminal work on program correctness. Finally, our library uses the algorithm of Lerner, Grove, and Chambers (2002), which enables compiler writers to compose very simple analyses and transformations in a way that achieves the same precision as complex, handwritten “super-analyses.” Our library is the workhorse of a new back end for the Glasgow Haskell Compiler.