ffc team mailing list archive

[Shawn Walker <walker@xxxxxxxxxxxxxxx>]

Actually, I'm a little worried about getting this initial trial to work. 
Here is the code that Poisson.h has in the inner most loop:

 // Number of operations to compute entry = 18
          A[j*6 + k] += ((Jinv_00*FE0_D10[ip][j] + 
Jinv_10*FE0_D01[ip][j])*(Jinv_00*FE0_D10[ip][k] + Jinv_10*FE0_D01[ip][k]) 
+ (Jinv_01*FE0_D10[ip][j] + 
Jinv_11*FE0_D01[ip][j])*(Jinv_01*FE0_D10[ip][k] + 
Jinv_11*FE0_D01[ip][k]))*W4[ip]*det;

My question is, can I just change Jinv_00 to Jinv_00[ip]?  That FE0_D10 
variable seems to be a special case structure.  Isn't the quadrature rule 
here built SPECIFICALLY for the case of a constant jacobian?  If I have a 
variable jacobian, won't that mess this up?

I can also just throw out the code that is there, and put in my own quad 
loop.  But that kind of defeats the purpose of this `proof-of-concept'.

- Shawn

On Mon, 22 Jun 2009, Anders Logg wrote:

> I think Shawn just wants to get the quadrature points so he can
> hand-modify the code for tabulate_tensor to handle higher-order mapped
> elements (to make a prototype for higher-order elements).
>
> Correct me if I'm wrong.
>
> In that case, we won't need to make them available since they are
> available to the form compiler when the code is being generated, but
> it would help if they were available in a comment in the generated code.
>
> --
> Anders
>
>
> On Mon, Jun 22, 2009 at 10:28:29PM +0200, Martin Sandve Alnæs wrote:
> > On Mon, Jun 22, 2009 at 9:10 PM, Kristian
> > Oelgaard<k.b.oelgaard@xxxxxxxxxx> wrote:
> > > Quoting Shawn Walker <walker@xxxxxxxxxxxxxxx>:
> > >
> > > > Actually, I am still a little confused.  Whenever you compute an integral
> > > > by quadrature, you need to evaluate the integrand at the quadrature
> > > > points.  Those are the points I need.
> > >
> > > If you have a quadrature element in your form
> > >
> > > element = FiniteElement("Quadrature", triangle, 2)
> > >
> > > you can call the
> > > tabulate_coordinates() function of the dof_map associated with the quadrature
> > > element. If you don't have a quadrature element there is no direct way of
> > > getting the quadrature points.
> > >
> > > > I just need the quad rule that is used in a particular case.  I see the
> > > > weights are defined in the tabulate_tensor routine.  Are the points listed
> > > > in a comment somewhere?
> > >
> > > No, currently not. But it might be a natural thing to have. I'll look into it.
> >
> > It could be an idea to make it available through UFC,
> > but then again each ufc::*_integral may use different
> > quadrature rules for different terms...
> > However, an exception can be thrown if there is no unique rule in use.
> >
> > Martin
> > _______________________________________________
> > FFC-dev mailing list
> > FFC-dev@xxxxxxxxxx
> > http://www.fenics.org/mailman/listinfo/ffc-dev