Compilation optimisante a l'aide de metaheuristiques (French text)

The construction of efficient compilers is very complex, since it has to face several optimization problems and it depends on the characteristics of the architecture of the machines for which they generate code. In this thesis we propose the use of the metaheuristic methods in compilation. Our thesis is that this makes it possible to obtain efficient code while simplifying the construction of the compilers. First, we show the viability to use the metaheuristic methods in compilation by replacing GNU C's register allocator by an allocator based on a genetic algorithm and by an allocator based on simulated annealing. Our allocators show a performance similar to GNU C. Afterwards, we define a unified approach to carry out the compilation using metaheuristic methods. This approach eliminates the weaknesses found in the traditional approaches. It allows to take account of the interdependencies between various problems present in the process of compilation; it does not make assumptions on the characteristics of the source programs so it permits to be more homogeneous on all of them. Finally, the approach allows the compiler to adapt easily to new changes to the architecture, since it is not very dependent on the characteristics of the machine. We test our approach on the problems of register allocation and instruction scheduling. We carry out an analysis of performance by comparing it with the more traditional approaches for these problems and we obtain profits on the speed of the generated code varying between −2% and 26%.