Created by: Xreki

resolve #4471 (closed) It is the basic implementation of sequence softmax operator, which calling Eigen to do all the computations. I am not sure how many times CUDA kernel will be launched in Eigen. I will try figure it out and optimize the implementation in next PR.