Merge pull request #5707 from qingqing01/cos_sim_op

Fix cos_sim_op in debug mode.

paddle/operators/cos_sim_op.h

@@ -132,7 +132,7 @@ class CosSimGradKernel : public framework::OpKernel<T> {
       // compute dy
       if (out_grad_y) {
         out_grad_y->mutable_data<T>(context.GetPlace());
-        auto dy = EigenMatrix<T>::Reshape(*out_grad_y, 1);
+        auto dy = EigenVector<T>::Flatten(*out_grad_y);
         auto grad = x / norm_prod_bcast - z_bcast * y_bcast / y_snorm_bcast;
         dy.device(place) = (dz_bcast * grad).sum(Eigen::array<int, 1>({{0}}));
       }