ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

This is a known problem. For higher-order forms and complicated terms,
the compile times are signifant. Just doing a wc on the generated .h
file should indicate why. It just takes long for gcc to compile a
function with tens of thousands of instructions, especially if it
tries to optimize.

As Garth points out, we are planning for FFC to be able to generate
quadrature instead of the tensor-based representation. This will help.

And there is also the option -f blas that generates less code at the
expense of a BLAS call but this is currently broken.

I'm working on a big rewrite/reorganization of FFC that will make it
simpler to maintain and add new options (like quadrature, blas etc).

/Anders


On Sat, Jan 20, 2007 at 12:35:17PM -0600, Andy Terrel wrote:
> Okay so switching the optimization flag from -O2 to -O0 is sped the 
> compilation time up by a factor of 100 but slowed the runtime by a 
> factor of .1 at least for order 5+, similar things happen for lower 
> orders but less dramatically so.  Which for me made many more tests 
> feasible.
> 
> Thanks,
> Andy
> 
> Andy Terrel wrote:
> > It is the DOLFIN executable that takes so long to compile.
> >
> > Garth N. Wells wrote:
> >   
> >> Andy Terrel wrote:
> >>     
> >>> Hello Guys,
> >>>
> >>> So comparing some of these Stokes methods, when I use a higher order 
> >>> iterated penalty method the compile time is killing me with 
> >>> ffc/dolfin.  As in it takes hours to compile and seconds to run.  
> >>>       
> >> Is it taking a long time to compile the form file, or to build a 
> >> DOLFIN executable?
> >>
> >> What do I do to
> >>     
> >>> speed these things up?  And do things run faster using pydolfin since 
> >>> python does the whole just in time thing?
> >>>
> >>>       
> >> If the form is changing, I wouldn't expect pyDOLFIN to help.
> >>
> >> Garth
> >>
> >>     
> >>> Thanks,
> >>> Andy
> >>> _______________________________________________
> >>> DOLFIN-dev mailing list
> >>> DOLFIN-dev@xxxxxxxxxx
> >>> http://www.fenics.org/mailman/listinfo/dolfin-dev
> >>>
> >>>       
> >>     
> >
> >
> >   
> 
>