Note: aluppold, doehlert, hfalk, ESD, WCC

Abstract: Optimizing embedded systems often boils down to solving complex combinatorial optimization problems. Integer-Linear Programming (ILP) turned out to be a powerful tool to solve these problems, as beyond traditional constraints, Boolean variables may be used to model complex logical expressions and conditionals. One of the key technical aspects is to be able to efficiently express these relations within the ILP. This paper presents formalized solutions for these issues, as well as an assessment of common ILP solvers. Additionally, the performance impact is illustrated using a compiler based cache aging optimization.