What't the arguments order of TVM generated cuda kernel?

Since we can’t know the cuda kernel args list, does TVM still keep the input and output args continuous, and by topological order?