Conclusions and Perspectives

Abstract

In this monograph we examined how to cope with prominent interactions in code motion in a systematic way. Common to all problems we studied was the dilemma that while, on the one hand, exploiting interactions offers a great potential for optimisations this, on the other hand, has to be paid by the increased conceptual and computational complexity of this process. Thus the design of an interacting code motion algorithm requires to capture as much of the optimisation potential as possible, while keeping the schedule as economically as possible. Our work was motivated by the observation that, as opposed to the data flow analysis based design of the component transformations, there is no foundation that deals with the effects of a system of interacting code motion transformations in concert. For a broad class of practically relevant problems in expression and assignment motion this monograph offered rigorous techniques for the design and analysis of aggressive algorithms in the presence of interactions.