Computing Same Block Relations for Relational Cache Analysis.

In contrast to the classical cache analysis of Ferdinand, the relational cache analysis does not rely on precise address information. Instead, it uses same block relations between memory accesses to predict cache hits. The relational data cache analysis can thus also predict cache hits if fully unrolling a loop is not feasible during analysis, for example due to high memory consumption or long computation time. This paper proposes a static analysis based on abstract interpretation which is able to compute same block relations for relational cache analysis.