Program transformations using temporal logic side conditions

Abstract

This paper describes an approach to program optimisation based on transformations, where temporal logic is used to specify side conditions, and strategies are created which expand the repertoire of transformations and provide a suitable level of abstraction. We demonstrate the power of this approach by developing a set of optimisations using our transformation language and showing how the transformations can be converted into a form which makes it easier to apply them, while maintaining trust in the resulting optimising steps. The approach is illustrated through a transformational case study where we apply several optimisations to a small program.

Specifying and Executing Optimizations for Parallel Programs

“…The basic approach of the PTRANS specification language is modeled after the TRANS language of Kalvala et al [4]: optimizations are specified as rewrites on program code in the form of control flow graphs, with side conditions given in Computation Tree Logic (CTL). Intuitively, the rewrite portion of an optimization expresses the transformation to be made, and the side condition characterizes the situations in which the optimization should be applied.…”

“…In this section we present the mathematical semantics of PTRANS, based on the semantics of TRANS by Kalvala et al [4]. The semantics of actions is given by a function A (σ , G ) that takes an action, a substitution (a partial map from metavariables to program objects), and a tCFG and returns the tCFG that results when the action is performed.…”

“…Our work builds on the TRANS approach of expressing optimizations as rewrites on control flow graphs with temporal logic side conditions due to Lacey et al [6] and Kalvala et al [4]. The most closely related tool is Cobalt [8], a system for specifying optimizations in a TRANS-like language.…”

“…Temporal logic formulae over program graphs allow us to simply and clearly state the conditions under which an optimization should be applied. PTRANS has both abstract mathematical semantics, derived from its predecessor language TRANS [4], and executable semantics that can be used by compiler designers to test and refine their transformations on actual program graphs. In this paper, we present the syntax and the abstract and executable semantics of PTRANS, and illustrate how it can be used to rapidly prototype and test compiler optimizations.…”

From Specification to Optimisation: An Architecture for Optimisation of Java Bytecode

“…1. Other optimisations specified in TRANS include lazy code motion, constant propagation, strength reduction, branch elimination, skip elimination, loop fusion, and lazy strength reduction; further details of these can be found in [4].…”