Multi-objective compiler optimization space exploration.

Modern compilers implement a wide range of code transformations that interact in a complex manner. This complicates the optimization of the generated code, because it is difficult to estimate the effect of a sequence of transformations. The core of the problem is that different code transformations have a different and sometimes opposite impact on different cost functions such as performance, power and energy consumption, memory usage, etc. The solution for this problem is to treat compiler optimization as a multi-objective optimization problem. This contrasts with current best practice, which uses heuristics or single-objective optimization. In this research project, we wish to speed up the simultaneous exploration of multiple criteria of the compiler optimization space. We will do this using scenario based exploration and model guided algorithms. We expect that this can lead to a speedup of two orders of magnitude. We will study the problem in the context of both static and dynamic compilers (virtual machines). This project uses joint expertise from three research labs: compilers and virtual machines (U. Gent), machine learning (K.U.Leuven), and scenario based optimizations (IMEC).

Keywords:Compiler optimization, Machine learning, Multi-objective optimization, Data mining

Disciplines:Computer hardware, Computer theory, Scientific computing, Other computer engineering, information technology and mathematical engineering, Applied mathematics in specific fields, Computer architecture and networks, Distributed computing, Information sciences, Information systems, Programming languages, Theoretical computer science, Visual computing, Other information and computing sciences