syfi team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Wed, Mar 25, 2009 at 11:41:29PM +0100, Martin Sandve Alnæs wrote:
> This should rather be fixed in dolfin and ffc instead.
> 
> 1) The module isn't needed, the module can always
> be obtained from compiled_form.__module__.
> 
> 2) We don't need to return the form_data either,
> since it can be obtained from ufl_form.form_data()
> (it's attached there the first time it is computed).
> 
> So we can just return the compiled module.
> 
> I want to do it this way:
> 
> single_compiled_form = jit(single_ufl_form)
> single_compiled_element = jit(single_ufl_element)
> compiled_object_list = jit(ufl_object_list)
> 
> where single_compiled_element is a
> tuple(single_compiled_finite_element, single_compiled_dof_map)
> 
> and compiled_object_list is an ordered list matching the
> ufl_object_list where each item can be either a single ufl form or element.

Sounds good. But does this need to be done *now* or can it wait until
we've moved the JIT compilation to a common module in UFC?

I'm hoping we can make a new release of FFC pretty soon with
experimental support for UFL and then make cleanups and fixes
for the next release (and remove .form support).

There will be a few things to do like using ufl.load_forms, using
dolfin_utils for generating DOLFIN code and changing the JIT
interface. It would be practical if those could wait until after the
release.

-- 
Anders

Attachment: signature.asc
Description: Digital signature