If $g(x)$ is smooth (assuming you mean at least $C^1$), then the iterated integrals are just Stieltjes integrals and
$$\int_s^t g(r)-g(s) dg (r)=\int_s^t (g(r)-g(s)) g'(r) dr$$
No rough path theory needed, because the path is smooth, lol.
Edit: for the updated question, no there is no nice relation as far as I know. For example take any path $g$ that is $C^{1+\alpha}$ for $\alpha<1/2$. I.e. $g$ is differentiable with $C^\alpha$ derivative. Then the signature of $g$ is just iterated Stieltjes integrals, but for $g'$ you need to postulate the iterated integrals which is of course non unique.
So no, there is no nice relationship.