Any benefit to using DLM for stochastic volatility model?

I’m looking at the part of the Stan manual on stochastic volatility models: 2.5 Stochastic volatility models | Stan User’s Guide

The unobserved log volatility process is an AR1 model. Even though the way this has been calculated here is quite optimised (eg through vectorisation), I wanted to check whether there might be a theoretical benefit to instead using gaussian_dlm_obs (see 22.7 Gaussian Dynamic Linear Models | Stan Functions Reference)

I’m not sure how this is coded internally, but I can imagine that this might be significantly more efficient (eg by using analytical derivatives, for instance).

I guess a downside is that this is the forward pass only, not the commensurate backward pass to get the smoothed values.