Note: aluppold, hfalk, ESD, emp2, tacle, WCC

Abstract: In hard real-time systems, each task has to provably finish its execution within its respective deadline. Compiler optimizations can be used to improve each task's timing behavior. However, current compilers do not consider tasks' deadlines and can therefore not be used to reliably optimize hard real-time systems with regard to its schedulability. We propose a compiler optimization framework based on Integer-Linear Programming which allows for schedulability aware code optimizations of hard real-time multitasking systems. We evaluate the framework using an instruction scratchpad optimization. The results show that our approach can be used to improve the schedulability of hard real-time systems significantly.