dolfin team mailing list archive

[A Navaei <axnavaei@xxxxxxxxxxxxxx>]

2009/3/23 Garth N. Wells <gnw20@xxxxxxxxx>:
>
>
> A Navaei wrote:
>>
>> 2009/3/23 Garth N. Wells <gnw20@xxxxxxxxx>:
>>>
>>> A Navaei wrote:
>>>>
>>>> The success of MTL4 based on generic meta-programming, arises the
>>>> question about re-visiting the efficiency of code-generation
>>>> approaches, including FFC. Given that FEM can particularly benefit
>>>> from major meta-programming characteristics, namely static
>>>> polymorphism and loop unrolling, MTL4 demonstrates that the
>>>> code-generation part can be much more efficiently replaced by inlining
>>>> performed at compile-time.
>>>>
>>> What FFC and other code generators do is a form of metaprogramming, so I
>>> don't see what point you're trying to make.
>>
>> As mentioned above, we are talking about compile-time implicit inline
>> code generation which is opposed to explicit code generation as done
>> by FFC -- refer to MTL publications.
>>
>
> No, you have addressed meta-programming which is the generation of code.

Why do you insist on that 'meta-programming' and 'explicit
code-generation' are the same concept? One could argue that the latter
could be achieved by 'meta-modeling', eg eclipse EMF, but the former
is more general and particularly involves compile-time inlining by
template meta-programming. In this context, as it is clear,
meta-programming is referred to c++ template meta-programming.

FFC
> does employ loop unrolling but it does not use static polymorphism. There is
> no fundamental reason why it could not also use static polymorphism (i.e.
> generate templated classes and functions).

What would be the point of using static polymorphism in FFC-generated
code followed by JIT? Why not using it directly?

>
> MTL4 does do some very neat things, and the performance for a number of
> operations is very impressive. However, if you look at the code you will see
> that for relatively simple operations it, and uBLAS, are very sophisticated.
> Advocating such an approach for differential equations without a proof of
> concept does not demonstrate an appreciation of the complexity of
> differential equations and variational methods over basic linear algebra
> operations.

It's the job of the DSL, in this case the form language, to hide the
complexity. The DSL can still live happily with the template
meta-programming approach.


-Ali

>
> Garth
>
>
>
>>
>> -Ali
>>
>>> Garth
>>>
>>>> Without having a concrete meta-programming implementation, it may be
>>>> impossible to predict how much performance one would gain compared to
>>>> FFC. However, MTL4 has been reported to be many times faster than
>>>> code-generation means such as ATLAS.
>>>>
>>>> Based on this, are there any specific benefits in FFC code-generation
>>>> which may not be covered by meta-programming?
>>>>
>>>>
>>>> -Ali
>>>> _______________________________________________
>>>> DOLFIN-dev mailing list
>>>> DOLFIN-dev@xxxxxxxxxx
>>>> http://www.fenics.org/mailman/listinfo/dolfin-dev
>>>
>>>
>
>
>