Analysis of irregular single-indexed array accesses and its applications in compiler optimizations

Abstract

Many compiler techniques require analysis of array subscripts to determine whether a transformation is legal. Traditional methods require the array subscript expressions to be expressed as closed form expressions of loop indices. Most methods further require the subscript expressions to be linear. However, in sparse/ irregular programs, closed-form expressions of array subscripts are not available. More powerful methods to analyze array subscripts are needed. Array accesses with no closed-form expressions available are called irregular array accesses. In real programs, many irregular array accesses are single-indexed. In this paper, we present techniques to analyze irregular single-indexed array accesses. We show that single-indexed array accesses often have properties that are useful in compiler analysis. We discuss how to use these properties to enhance compiler optimizations. We also demonstrate the application of these techniques in three real-life programs to exploit more implicit parallelism.