Task Decomposition and Parallelization Planning for Automotive Power-Train Applications

Since both the industry and the market have been driven by the multi-core processors, to parallelize the applications in the automotive industry is on huge demand. Two steps are required to parallelize a legacy program: parallelism discovery and parallelization planning. To discover the parallelism of a program is to identify the code regions where multiple procedures/ functions can be executed simultaneously, while parallelization planning is to find an optimal solution of assigning tasks on multi-cores based on the discovered parallelism. How to automate the parallelization in the Power-train domain remains a grand challenge due to the complexity of the program and the dynamics of the runtime environment. Many aspects should be considered including the speedup, computing resource bound, workload balance, etc. Considering all the above aspects, we used a directed acyclic graph to represent the decomposed program, then take the parallelization planning as a multi-objective optimization problem, where a Cobyla algorithm is deployed to search for the optimal solution by evaluating different parameters. We have tested our approach on the periodic tasks in the Power-train applications to validate its feasibility and efficiency.