transpose_mkldnn code change to meet Paddle standards (#22591)

paddle/fluid/operators/mkldnn/transpose_mkldnn_op.cc

@@ -40,7 +40,8 @@ class TransposeMKLDNNOpKernel : public paddle::framework::OpKernel<T> {
     const T* input_data = input->data<T>();
 
     if (ndims == 1) {
-      output->ShareDataWith(*input);
+      framework::TensorCopy(*input, input->place(), output);
+      output->set_format(input->format());
       return;
     }
 
@@ -85,7 +86,8 @@ class TransposeMKLDNNGradOpKernel : public paddle::framework::OpKernel<T> {
     std::vector<int> reversed_axis(axis);
     int ndims = axis.size();
     if (ndims == 1) {
-      x_grad->ShareDataWith(*out_grad);
+      framework::TensorCopy(*out_grad, out_grad->place(), x_grad);
+      x_grad->set_format(out_grad->format());
       return;
     }