dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Wed, Oct 05, 2005 at 03:41:33PM +0200, Garth N. Wells wrote:

> For higher-order (p > 1) complicated forms, I'm seeing big improvements in
> compile time when using BLAS mode.

Great!

> Run-time performance appears very similar.

ok. I'm about to start with some more extensive benchmarking. I'll
report the results later.

> I can't test big problems because BLAS mode has *huge* memory
> requirements when using higher order elements. For Poisson equation
> with p = 10, I can't use any more than 35x35 elements without having
> a PETSc memory error.

p = 10 is pretty high... you can add a line

   blas.disp();

somewhere in your eval() function in the form generated by FFC and see
how many entries you have in the matrix that gets shipped off to BLAS.

It sounds strange that you get a PETSc memory error. Degree 10 on
tetrahedra has 286 degrees of freedom, so rougly 286*35*35 = 350350
degrees of freedom total. That should be no problem for PETSc (if you
have sufficient memory).

> Separate issue - with more complicated forms, I have trouble with anything
> around p > 2 due to the memory requirements of ffc. Are there any simple
> solutions to this?

I wasn't aware of this. With degree 10 Poisson, memory usage stays
around 30MB. Can you give me an example (as simple as possible) that
requires a lot of memory?

/Anders