An Experimental Framework for Whole System Optimization

It have been always a need to optimize the systems, specially on embedded environments that are high constrained in terms of memory footprint or power consumption. The current approach to optimize the systems is to perform manual tailoring and apply automatic optimizations at compile-time, avoiding the high cost of develop optimal components from scratch. Nevertheless, the manual tailoring of the system is still hard and error-prone because the most useful optimiza-tions remain hidden to the automatic tools. The reason is the current tools have a lack of a global view of system (GSV). The new approach presented takes into account the GSV and allows to perform more aggressive optimizations at link-time on the system components, reducing the manual customization while keeping reliability. We apply optimizations such as dead code elimination and constant propa-gation across the components of an embedded system. The results show an up to 40 % of code reduction on the system