dolfin team mailing list archive

[Johan Jansson <jjan@xxxxxxxxxxx>]

On Tue, Jan 16, 2007 at 09:04:31PM +0100, Anders Logg wrote:
> Looks good, but this might not be needed with the UFC-based assembly.
> 
> The current BilinearForm interface for evaluation of the element
> tensor looks like this:
> 
>     void eval(real block[],
>               const AffineMap& map,
>               real det) const;
> 
> but the corresponding UFC interface looks like this:
> 
>     void tabulate_tensor(double* A,
>                          const double * const * w,
>                          const ufc::cell& c) const;
> 
> A is the same as block.
> 
> w is a list of expansion coefficients for the functions in the form,
> and corresponds to the member data c in BilinearForm.
> 
> The AffineMap and determinant are no longer arguments to the
> tabulate_tensor (eval) function, since they may not be needed, so the
> generated code will need to compute the affine map and the determinant
> when they are needed.
> 
> Perhaps the AffineMap will be needed for other things, but it won't be
> needed for the generated code.
> 
> /Anders
> 

Ok good, I was just about to write a mail asking about this. Right now
I'm looking at evaluation of basis functions on physical cells. The
plan is to define a few common basis functions on the reference cell
and just map them. But all of this will be obsoleted by UFC, as you
just verified. It's probably still worth doing (because it's needed
right now), but it's always good to know what's planned ahead.

  Johan