Automatic and Explicit Parallelization Approaches for Mathematical Simulation Models

Sammanfattning: The move from single core and processor systems to multi-core and many-processors systemscomes with the requirement of implementing computations in a way that can utilizethese multiple units eciently. This task of writing ecient multi-threaded algorithmswill not be possible with out improving programming languages and compilers to providethe mechanisms to do so. Computer aided mathematical modeling and simulationis one of the most computationally intensive areas of computer science. Even simpli-ed models of physical systems can impose a considerable amount of computational loadon the processors at hand. Being able to take advantage of the potential computationpower provided by multi-core systems is vital in this area of application. This thesis triesto address how we can take advantage of the potential computation power provided bythese modern processors to improve the performance of simulations. The work presentsimprovements for the Modelica modeling language and the OpenModelica compiler.Two approaches of utilizing the computational power provided by modern multi-corearchitectures are presented in this thesis: Automatic and Explicit parallelization. Therst approach presents the process of extracting and utilizing potential parallelism fromequation systems in an automatic way with out any need for extra eort from the modelers/programmers side. The thesis explains improvements made to the OpenModelicacompiler and presents the accompanying task systems library for ecient representation,clustering, scheduling proling and executing complex equation/task systems with heavydependencies. The Explicit parallelization approach explains the process of utilizing parallelismwith the help of the modeler or programmer. New programming constructs havebeen introduced to the Modelica language in order to enable modelers write parallelizedcode. the OpenModelica compiler has been improved accordingly to recognize and utilizethe information from this new algorithmic constructs and generate parallel code toimprove the performance of computations.