Benchmark framework of compute-in-memory based accelerators for deep neural network (inference engine focused)