Multigrain Parallelization for Model-Based Design Applications Using the OSCAR Compiler

Model-based design is a very popular software development method for developing a wide variety of embedded applications such as automotive systems, aircraft systems, and medical systems. Model-based design tools like MATLAB/Simulink typically allow engineers to graphically build models consisting of connected blocks for the purpose of reducing development time. These tools also support automatic C code generation from models with a special tool such as Embedded Coder to map models onto various kinds of embedded CPUs. Since embedded systems require real-time processing, the use of multi-core CPUs poses more opportunities for accelerating program execution to satisfy the real-time constraints. While prior approaches exploit parallelism among blocks by inspecting MATLAB/Simulink models, this may lose an opportunity for fully exploiting parallelism of the whole program because models potentially have parallelism within a block. To unlock this limitation, this paper presents an automatic parallelization technique for auto-generated C code developed by MATLAB/Simulink with Embedded Coder. Specifically, this work 1) exploits multi-level parallelism including inter-block and intra-block parallelism by analyzing the auto-generated C code, and 2) performs static scheduling to reduce dynamic overheads as much as possible. Also, this paper proposes an automatic profiling framework for the auto-generated code for enhancing static scheduling, which leads to improving the performance of MATLAB/Simulink applications. Performance evaluation shows 4.21 times speedup with six processor cores on Intel Xeon X5670 and 3.38 times speedup with four processor cores on ARM Cortex-A15 compared with uniprocessor execution for a road tracking application.