Update to AD component of quasi-static Finite-strain compressible elasticity

[jppelteret]

This modernises the AD data structures used in this code-gallery example.

[jppelteret]

Currently WIP because I would like to generalise this to all ad_type_codes that we support (i.e. enable ADOL-C support and choose the AD library in the parameter file).

[davydden]

I started applying this to the code derived from the code gallery and indeed, we certainly need a way to go from an arbitrary (AD) NumberType to the enumerator, and thus as you said, we need something like none for non-AD types.

EDIT: I think if works for current AD number types, that would actually already be enough, at least in my case, as far as I can tell. Let me se if I can add this tomorrow...

[jppelteret]

I actually like the idea of an AD::NumberTypes::None. It would have simplified some of my own code, so I think I'd still like to implement it. It shouldn't be too much work.

[davydden]

sure, I agree, it will be useful regardless. What I meant is that even without None the link from AD numbers to enumerators can help.

[jppelteret]

For the record, Differentiation::AD::ADNumberTraits<ADNumberType>::type_code defines this backwards link.

[davydden]

nice 👍 Did not know about that. Thanks for the link.

[davydden]

Currently WIP because I would like to generalise this to all ad_type_codes that we support (i.e. enable ADOL-C support and choose the AD library in the parameter file).

I played around and got the support for any AD numbers working (seems so at least) in my derived code using AD::NumberTraits::n_supported_derivative_levels. However this constraints the choice between residual and energy AD purely based upon n_supported_derivative_levels meaning one can't use adolc_taped for residual linearization. That's probably not how we should recommend using ADHelper, but that's how I got this setup for now. Just my 2c.

[jppelteret]

However this constraints the choice between residual and energy AD purely based upon n_supported_derivative_levels meaning one can't use adolc_taped for residual linearization.

You could get around this with a more specific check, for example:

enable_if<ADNumberTraits::is_taped || (ADNumberTraits::is_tapeless && ADNumberTraits::n_supported_derivative_levels >= 2)>

I think that this should work in your case.