ffc team mailing list archive

[Harish Narayanan <harish.mlists@xxxxxxxxx>]

On 12/8/11 11:07 AM, Kristian Ølgaard wrote:
> On 8 December 2011 10:58, Harish Narayanan <harish.mlists@xxxxxxxxx> wrote:
>> On 12/8/11 10:53 AM, Anders Logg wrote:
>>> On Thu, Dec 08, 2011 at 10:46:20AM +0100, Harish Narayanan wrote:
>>>> Dear all,
>>>>
>>>> I am trying to run some problems which involve somewhat complex
>>>> variational forms which FFC takes ages (O(1 day)) to compile. Is there
>>>> any clear documentation on the strategies one should apply to speed
>>>> things up in general? I am interested both in appropriate FFC parameters
>>>> and tips on how to split a form into pieces so that I don't end up
>>>> recompiling everything every time I make small changes.
>>>
>>> Did you try setting the number of quadrature points? FFC is likely
>>> using quite a few points to integrate your form and it's probably
>>> overkill.
>>
>> Is that quadrature_order or quadrature_degree? I have unsystematically
>> tried a few things which haven't been successful, so I thought I would
>> ask for documentation.
> 
> quadrature_degree
> 
> see 'man ffc'
> 
> There is no documentation on which strategies to use wrt splitting up
> forms, and I don't really have any good advice from the top of my
> head.
> (I assume you use PyDOLFIN)
> 
> Using no optimisation for the FFC compiler will obviously result in
> the fastest compile time.

I will try removing all optimisation and reducing the quadrature_degree.

Thanks!
Harish