Parallelization Planning for Legacy Programs: A Multi-Objective Genetic Algorithm Approach

Parallelization of legacy programs is on huge demand as multi-core platforms are pervasively adopted in many sectors of industry. In order to parallelize a legacy program, two steps must be done: parallelism discovery and parallelization planning. Parallelism discovery is the process of identifying code regions where have exploitable parallelism. And for parallelization planning, many aspects should be considered including but not limited to the speedup, core load balancing, communication costs, etc. In this work, we model the parallelization planning as a multi-objective optimization problem. A genetic algorithm is developed to solve this multi-objective optimization problem by evaluating the solution set taking consideration of all the above aspects. We have tested our approach on the real industrial applications to validate its feasibility and efficiency.