Genetic Algorithm Based Parallelization Planning for Legacy Real-Time Embedded Programs

Multicore platforms are pervasively deployed in many different sectors of industry. Hence, it is appealing to accelerate the execution through adapting the sequential programs to the underlying architecture to efficiently utilize the hardware resources, e.g., the multi-cores. However, the parallelization of legacy sequential programs remains a grand challenge due to the complexity of the program analysis and dynamics of the runtime environment. This paper focuses on parallelization planning in that the best parallelization candidates would be determined after the parallelism discovery in the target large sequential programs. In this endeavor, a genetic algorithm based method is deployed to help find an optimal solution considering different aspects from the task decomposition to solution evaluation while achieving the maximized speedup. We have experimented the proposed approach on industrial real time embedded application to reveal excellent speedup results.