Performance modeling of embedded applications with zero architectural knowledge

Performance estimation is a key step in the development of an embedded system. Normally, the performance evaluation is performed using a simulator or a performance mathematical model of the target architecture. However, both these approaches are usually based on the knowledge of the architectural details of the target. In this paper we present a methodology for automatically building an analytical model to estimate the performance of an application on a generic processor without requiring any information about the processor architecture but the one provided by the GNU GCC Intermediate Representation. The proposed methodology exploits the linear regression technique based on an application analysis performed on the Register Transfer Level internal representation of the GNU GCC compiler. The benefits of working with this type of model and with this intermediate representation are three: we take into account most of the compiler optimizations, we implicitly consider some architectural characteristics of the target processor and we can easily estimate the performance of portions of the specification. We validate our approach by evaluating with cross-validation technique the accuracy and the generality of the performance models built for the ARM926EJ-S and the LEON3 processors.