Parallel Code Generation in MathModelica / An Object Oriented Component Based Simulation Environment

Modelica is an a-causal, equation based, object oriented modeling language for modeling and efficient simulation of large and complex multi domain systems. The Modelica language, with its strong software component model, makes it possible to use visual component programming, where large complex physical systems can be modeled and composed in a graphical way. One tool with support for both graphical modeling, textual programming and simulation is MathModelica. To deal with growing complexity of modeled systems in the Modelica language, the need for parallelization becomes increasingly important in order to keep simulation time within reasonable limits. The first step in Modelica compilation results in an Ordinary Differential Equation system or a Differential Algebraic Equation system, depending on the specific Modelica model. The Modelica compiler typically performs optimizations on this system of equations to reduce its size. The optimized code consists of simple arithmetic operations, assignments, and function calls. This paper presents an automatic parallelization tool that builds a task graph from the optimized sequential code produced by a commercial Modelica compiler. Various scheduling algorithms have been implemented, as well as specific enhancements to cluster nodes for better computation/communication tradeoff. Finally, the tool generates simulation code, in a master-slave fashion, using MPI.