Contents

For many software projects, it is desirable to use the features of an object-oriented programming language such as C++ to write maintainable code, to implement clean interfaces, and to facilitate code reuse. C++ virtual functions are a central building block for object-oriented software archi-tectures. In certain use cases like operating system kernels and especially in microkernels extremely ecient code is required. However, such mechanisms impose too much run-time overhead, making the use of object-oriented code too impractical. This is one of the classic reasons against the use of object-oriented languages for building performance-critical systems. This study thesis describes an optimization technique that allows the use of C++ class hierarchies with virtual functions while producing ecient code. This is achieved by transforming the source code to collapse any specied class hierarchy into one at class. Semantic equivalence is maintained as far as necessary, in a limited environment. The technique is implemented in a source-to-source transformation tool which this paper discusses in detail. Some evaluation shows that the tool produces optimized code which is more ecient, without producing signicant compile-time overhead