Implementation of independent loops in high performance C

In parallel programming, loop computation is one of the most expensive operations in terms of execution duration. In order to solve this problem their parallelization is compulsory. One form of parallelization is for a compiler to automatically produce a parallel SPMD (Single Program Multiple Data) program. These compilation techniques encompass both compile-time and run-time resolution methods.This thesis presents a specific case of the run-time resolution of the loops, the independent loops. These loops do not have any interactions between their iterations, hence they have a great potential for parallelism. This work presents an experimental method which has been developed in the EPPP ("Environment de programmation parallele portable") project at the Centre de Recherche Informatique de Montreal.The development of the independent loops uses active messages, based on a particular buffering method, and uses also multi-threading. The implementation of the active messages is done on several types of communications libraries, in order to ensure the portability of the EPPP environment. The use of buffers considerably decreases the communications required during the calculation of the independent loops. Finally, the multi-threading system is used to suppress the idle-time appearing when a processor is waiting for some data issued by a remote processor.This thesis provides a description of the techniques and experimental data to show their applicability