Loop squashing transformations for amorphous slicing

Abstract

Program slicing is a source code extraction technique that can be used to support reverse engineering by automatically extracting executable subprograms that preserve some aspect of the original program's semantics. Although minimal slices are not generally computable, safe approximate algorithms can be used to good effect. However, the precision of such slicing algorithms is a major factor in determining the value of slicing for reverse engineering. Amorphous slicing has been proposed as a way of reducing the size of a slice. Amorphous slices preserve the aspect of semantic interest, but not the syntax that denotes it, making them generally smaller than their syntactically restricted counterparts. Amorphous slicing is suitable for many reverse engineering applications, since reverse engineering typically abandons the existing syntax to facilitate structural improvements. Previous work on amorphous slicing has not attempted to exploit its potential to apply loop-squashing transformations. This paper presents an algorithm for amorphous slicing of loops, which identifies induction variables, transformation rule templates and iteration-determining compile-time expressions. The algorithm uses these to squash certain loops into conditional assignments. The paper also presents an inductive proof of the rule templates and illustrates the application of the algorithm with a detailed example of loop squashing.