ffc team mailing list archive

["Martin Sandve Alnæs" <martinal@xxxxxxxxx>]

On Fri, Dec 12, 2008 at 12:32 PM, Anders Logg <logg@xxxxxxxxx> wrote:
> On Fri, Dec 12, 2008 at 11:59:35AM +0100, Kristian Oelgaard wrote:
>> Quoting Martin Sandve Alnæs <martinal@xxxxxxxxx>:
>>
>> > In principle UFL isn't concerned with integration method,
>> > since this is the main difference between the form compilers.
>>
>> I think it would make sense if UFL would compute the degree of a monomial, and
>> attach it as a member. Then the compilers could just ask for this info and
>> generate the proper integration rule (potentially user specified)
>
> UFL uses a different representation. I will implement an algorithm for
> extracting the monomial representation (like we have now in FFC) from
> the UFL representation (which is an AST - Abstract Syntax Tree with
> either an operator or a terminal symbol at each node) which will be
> useful to compute the tensor representation.
>
> The quadrature code generation should work on the AST instead of the
> monomial representation and I imagine it will need to first extract
> all terms and then find the degree for each term. I suspect Martin
> already has some nice utilities for this.

I imagine that we can share quite a lot of code for the quadrature
code generators.
But I don't have any code for the degree computation yet.

The current status is this:
I'm currently working on getting the general compiler framework up to date,
the whole chain ufl -> sfc -> ufc -> dolfin (C++). This will hopefully work
at the end of today.

Then I'll be implementing one demo at a time, beginning with projection
and poisson, and handle remaining bugs and issues as I hit them.
When some basic forms are working, the AD stuff needs work.

Martin