As transistors size and power limitations stroke computer industry, hardware parallelism arose as the solution, bringing old forgotten problems back into equation to solve the existing limitations of current parallel technologies. Compilers regain focus by being the most relevant puzzle piece in the quest for the expected computer performance improvements predicted by Moores law no longer possible without parallelism. Parallel research is mainly focused in either the language or architectural aspects, not really giving the needed attention to compiler problems, being the reason for the weak compiler support by many parallel languages or architectures, not allowing to exploit performance to the best. This thesis addresses these problems by presenting: Erbium, a low level streaming data-flow language supporting multiple producer and consumer task communication; a very efficient runtime implementation for x86 architectures also addressing other types of architectures; a compiler integration of the language as an intermediate representation in GCC; a study of the language primitives dependencies, allowing compilers to further optimise the Erbium code not only through specific parallel optimisations but also through traditional compiler optimisations, such as partial redundancy elimination and dead code elimination.

Frappée par les rendements décroissants de la performance séquentielle et les limitations thermiques, l’industrie des microprocesseurs s’est tournée résolument vers les multiprocesseurs sur puce. Ce mouvement a ramené des problèmes anciens et difficiles sous les feux de l’actualité du développement logiciel. Les compilateurs sont l’une des pièces maitresses du puzzle permettant de poursuivre la traduction de la loi de Moore en gains de performances effectifs, gains inaccessibles sans exploiter le parallélisme de threads. Pourtant, la recherche sur les systèmes parallèles s’est concentrée sur les aspects langage et architecture, et le potentiel reste énorme en termes de compilation de programmes parallèles, d’optimisation et d’adaptation de programmes parallèles pour exploiter efficacement le matériel. Cette thèse relève ces défis en présentant Erbium, un langage de bas niveau fondé sur le traitement de flots de données, et mettant en œuvre des communications multi-producteur multi-consommateur ; un exécutif parallèle très efficace pour les architectures x86 et des variantes pour d’autres types d’architectures ; un schéma d’intégration du langage dans un compilateur illustré en tant que représentation intermédiaire dans GCC ; une étude des primitives du langage et de leurs dépendances permettant aux compilateurs d’optimiser des programmes Erbium à l’aide de transformations spécifiques aux programmes parallèles, et également à travers des formes généralisées d’optimisations classiques, telles que l’élimination de redondances partielles et l’élimination de code mort. / As transistors size and power limitations stroke computer industry, hardware parallelism arose as the solution, bringing old forgotten problems back into equation to solve the existing limitations of current parallel technologies. Compilers regain focus by being the most relevant puzzle piece in the quest for the expected computer performance improvements predicted by Moores law no longer possible without parallelism. Parallel research is mainly focused in either the language or architectural aspects, not really giving the needed attention to compiler problems, being the reason for the weak compiler support by many parallel languages or architectures, not allowing to exploit performance to the best. This thesis addresses these problems by presenting: Erbium, a low level streaming data-flow language supporting multiple producer and consumer task communication; a very efficient runtime implementation for x86 architectures also addressing other types of architectures; a compiler integration of the language as an intermediate representation in GCC; a study of the language primitives dependencies, allowing compilers to further optimise the Erbium code not only through specific parallel optimisations but also through traditional compiler optimisations, such as partial redundancy elimination and dead code elimination.