Efficient Design Space Exploration of OpenCL Kernels for FPGA Targets Using Black Box Optimization

Nowadays, many industries are in favor of using intelligent design space exploration as opposed to brute-force analysis. In applications, the is defined by multiple variables and their interactions. Although analysis very simple, it rarely scalable when number system increases. With rising complexity hardware designs, more approaches needed explore options. This paper proposes smart meta-heuristic search algorithms such Grey Wolf Optimization (GWO) conjunction with Bayesian (BO) solve this problem. We show that we can further reduce effort a surrogate model created based on novel hybrid GWO-BO method. The useful abstraction detect functional physical inter-dependencies order accurately predict its performance (e.g. throughput or latency). evaluate our methodology produce competitive results find best maximize system. Finally, compare previous statistical heuristic methods proposed literature method always performs better than other considered methods.