Hi,

To add features to the tensor representation-based code generation, you
need to make changes to FFC.

More specifically, my recommendations would be to:

1. Read chapter X in the FEniCS book (can be downloaded from web page).

2. Dig into the code in FFC in the subdirectory ffc/tensor

3. Make changes to ffc/tensor/monomialtransformation.py and possibly other
files.

4. Create a simple test case (a UFL file).

5. Don't be afraid to experiment with <change Python code>  - <compile form
with ffc foo.ufl> - <look at C++ code>.

6. Good luck. :-)

--
Anders


tis 16 juni 2015 kl 16:26 skrev Steven Vandekerckhove <
[email protected]>:

> Dear all,
>
> I'm using a DG discretization to solve different kinds of wave equations
> in time domain.
> I would like to use faster assembly for these systems. Garth suggested
> using
> parameters['form_compiler']['representation'] = 'tensor'
> in stead of the default
> parameters['form_compiler']['representation'] = 'quadrature'
>
> Unfortunatly the tensor representation appears not to support Facet
> Normals:
> "ffc.tensor.monomialextraction.MonomialException: No handler defined for
> terminal FacetNormal."
>
> I would be happy to invest some time in adding the necessary code to
> make this possible (since I really need the speed-up),
> is there anyone who could get me started (Give a rough idea of what has
> to be done, and what scripts I should look at)?
>
> Thanks in advance,
> Steven
>
> P.s. I have some experience with making adjustments to dolfin
> _______________________________________________
> fenics mailing list
> [email protected]
> http://fenicsproject.org/mailman/listinfo/fenics
>
_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics

Reply via email to