ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Sun, Jan 31, 2010 at 04:13:26PM -0800, Johan Hake wrote:
> On Sunday 31 January 2010 15:37:41 Garth N. Wells wrote:
> > Anders Logg wrote:
> > > On Sun, Jan 31, 2010 at 05:51:37PM +0000, Garth N. Wells wrote:
> > >> Anders Logg wrote:
> > >>> On Sun, Jan 31, 2010 at 04:57:36PM +0000, Garth N. Wells wrote:
> > >>>> Nearly all the DOLFIN demos are working again. The only remaining
> > >>>> issue appears to be the automatic selection of the cell type for
> > >>>> Expressions from the Python interface, see for example the Python
> > >>>> demo. If 'cell=triangle' is added to the Expressions in the Python
> > >>>> demo, it all looks good.
> > >>>>
> > >>>> Garth
> > >>>
> > >>> Should be fixed now.
> > >>>
> > >>> Which demos were failing? I checked Poisson 1D which works now.
> > >>
> > >> Poisson (2D) and a few others (while Poisson is broken, I don't take
> > >> much notice of the others which are broken), and some of my solvers.
> > >> Seems ok now.
> > >>
> > >> It would be useful if the representation would be printed to the screen
> > >> when using JIT.
> > >
> > > You can print it by decreasing the debug level. Just set the
> > > "log_level" option to INFO.
> > >
> > >> How do I control FFC options from PyDOLFIN?
> > >
> > > By passing an optional 'options' argument to assemble() or
> > > VariationalProblem:
> > >
> > >   A = assemble(a, options={"representation": "quadrature",
> > >                            "log_level": INFO})
> > >
> > > This was broken but it works again now. VariationalProblem didn't have
> > > an options argument before but I have added it. I have also renamed
> > > form_compiler_options --> options since it's shorter and there is no
> > > other options dictionary.
> >
> > Setting FFC parameters seems a but clunky. How can one set global
> > parameters (along the lines of 'parameters.optimize = True', which does
> > work at the moment), and how does one know what keys can go into the
> > options dictionary?
>
> I think this is done by design as PyDOLFIN should be able to use different
> form_compilers. The option dict should just be passed to the form_compiler
> using the options argument.
>
> It seems that parameter.optimize only deals with caching of tensors and that,
>
>    parameters.optimize_form
>    parameters.optimize_use_dof_map_cache
>
> do not do anything more?
>
> We could force the form_compiler to define a more verbal option interface,
> explaining the meaning of the different options, which then could be reached
> from PyDOLFIN?

I'll see what I can do. It should be fairly easy to integrate the form
compiler options with the DOLFIN parameter system (without DOLFIN
needing to know about specific form compiler options).

But one thing that bugs me is that it's called 'options' in FFC and
'parameters' in DOLFIN. Is there some logic behind this or should we
try to make it uniform?

--
Anders

Attachment: signature.asc
Description: Digital signature