[DNNL] Check for distributing training example and try running on MKL-DNN engine
Created by: jczaja
Goal is to evaluate PaddlePaddle distributed training by finding and reproducing some multi-node example and then explore what problems are to have it running using DNNL