Pull BasisIsCollocated helper to interface level

[jeremylt]

This will help add this to gen/shared cleanly

[jrwrigh]

Also need to update the release notes.

[jrwrigh]

What if the 1D basis and quadrature points aren't assigned? Is interp_1d and the like always defined?

[jeremylt]

Those are always assigned for a tensor basis. Impossible to make a tensor basis without them

[jrwrigh]

Sure. I'm thinking about this for non-tensor basis, which can have collocated a collocated basis-quadrature pair.

[jeremylt]

currently not something we've implemented so we'd have to revisit when we find a use case

[jrwrigh]

My point is that a function called CeedBasisIsCollocated() should say whether the basis is collocated rather than whether it's collocated and tensor. I know this is a backend function, but it should still be clear what it does.

This would also make it more clear which cases are implemented and which are not in the backend code. Something like

CeedBasisIsTensor(basis, &is_tensor);
CeedBasisIsCollocated(basis, &is_collocated);

if (is_tensor && is_collocated) {
... // optimized code path (e.g. CeedBasisApply == memcpy)
} else {
... // normal code path
}

is more clear than if we relied on CeedBasisIsCollocated() to be the entire determination of whether to take a more optimized code path.

[jeremylt]

Sure, but we don't currently have any non-tensor collocated sort of logic, so I don't want to spend too much effort trying to get ahead of use cases that don't currently exist

[jrwrigh]

Fair. The limitation should be documented then.

Checking a non-tensor basis is not implemented and will always return false