ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

Garth N. Wells wrote:
>
> Kristian Oelgaard wrote:
>> Quoting Martin Sandve Alnæs <martinal@xxxxxxxxx>:
>>
>>> Sorry, didn't look at the generated code. FFC is generating quadrature
>>> code, I thought the default argument quadrature_points=False to jit
>>> options meant tensor representation. How do I choose tensor
>>> representation with jit?
>> I thought FFC was using tensor representation by default? The 
quadrature_points
>> option can be used to specify how many quadrature points are used if one
>> compiles using the quadrature representation.
>>
>
> Anders might have set quadrature as default for JIT as it compiles
> faster than the quadrature representation.

Not that I know of, but we should keep this in mind if we want to
reduce the compile times.

I just double-checked and we generate code by tensor-representation in
the jit compiler (if not otherwise specified).

Martin, are you using any special options to jit()?

And are you sure you get quadrature code?

/Anders


> Garth
>
>
>> Kristian
>>
>>>
>>>
>>> 2007/10/16, Martin Sandve Alnæs <martinal@xxxxxxxxx>:
>>>> 2007/10/16, Garth N. Wells <gnw20@xxxxxxxxx>:
>>>>> Martin Sandve Alnæs wrote:
>>>>>> In the FFC manual it says that --optimize is currently ignored, does
>>>>>> this mean that optimization isn't working at all at the moment?
>>>>>>
>>>>> It means that FErari isn't being used.
>>>> When will it be reimplemented then?
>>>>
>>>>>> We're doing benchmarks of sfc vs ffc vs dealII vs diffpack, and ffc
>>>>>> isn't doing as well as I expected.
>>>>>>
>>>>> Are you looking at just assembly?
>>>>>
>>>>> Garth
>>>> Just computing the element tensor. (We'll do full assembly too later.)
>>>>
>>>>
>>> _______________________________________________
>>> FFC-dev mailing list
>>> FFC-dev@xxxxxxxxxx
>>> http://www.fenics.org/mailman/listinfo/ffc-dev
>>>
>>
>> _______________________________________________
>> FFC-dev mailing list
>> FFC-dev@xxxxxxxxxx
>> http://www.fenics.org/mailman/listinfo/ffc-dev
>>
>
> _______________________________________________
> FFC-dev mailing list
> FFC-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/ffc-dev