dolfin team mailing list archive

["Alessio Quaglino" <alessio@xxxxxx>]

>
>
> Alessio Quaglino wrote:
>>> Alessio Quaglino wrote:
>>>>> Alessio Quaglino wrote:
>>>>>>>> - I don't know how to set the variable DOLFIN_VERSION present in
>>>>>>>> init.cpp
>>>>>>> DOLFIN_VERSION is a #define (see
>>>>>>> src/kernel/main/dolfin/constants.h).
>>>>>> Yes sorry, the compiler actually says that the PACKAGE_VERSION
>>>>>> variable
>>>>>> is
>>>>>> undefined.
>>>>>>
>>>>>> I managed to compile and link everything (using default BLAS
>>>>>> libraries
>>>>>> and
>>>>>> I still haven't checked if UMFPACK works), however I still have
>>>>>> quite
>>>>>> big
>>>>>> problems since my code runs very slowly (compared to linux) and
>>>>>> almost
>>>>>> gets stuck in the lump() routine because it takes ages.
>>>>>>
>>>>> Are you using uBLAS or PETSc? If you're using uBLAS, check the
>>>>> compiler
>>>>> flags and make sure that you have -DNDEBUG. Otherwise, anything to do
>>>>> with uBLAS will be dead slow.
>>>> Yes, thanks. It seems VS had the similar flag -NDEBUG which helped
>>>> quite
>>>> a
>>>> lot. I was also wondering if I could use some optimized BLAS like
>>>> ATLAS.
>>> DOLFIN doesn't use BLAS, but it may be used by libraries which DOLFIN
>>> uses, such as PETSc and UMFPACK. uBLAS doesn't use BLAS.
>>
>> Just to be sure, since understanding how to use PETSc in VS will be
>> cumbersome and time-consuming, if I switch to PETSc I can completely
>> forget about uBLAS and get a faster implementation?
>
> No and no. DOLFIN is dependent on UMFPACK, and PETSc is not necessarily
> faster. The reason for using PETSc is to permit parallel computation and
> access to a wide range of solvers and preconditioners.

I still want to use UMFPACK + ATLAS, but my bottleneck are some
matrix-matrix multiplication taking 60% of the runtime execution because
performed by uBLAS (see this becnhmark
http://www.crystalclearsoftware.com/cgi-bin/boost_wiki/wiki.pl?Frequently_Asked_Questions_Using_UBLAS).
Also, the compiling takes ages, and I want to understand if this depends
on uBLAS or Dolfin. That's why I want to try PETSc.

Alessio


> Garth
>
>
> Are there any
>> drawbacks coming from this choice (I suppose so, otherwise one would
>> never
>> use uBLAS)?
>>
>> Thanks,
>> Alessio
>>
>>
>>> Garth
>>>
>>>> Is ATLAS independent from BOOST or do I need to have both installed?
>>>>
>>>> Alessio
>>>>
>>>>
>>>>> Garth
>>>>>
>>>>>>> It would be helpful if you submitted your fixes as an hg bundle (or
>>>>>>> a
>>>>>>> series of small bundles, one for each set of fixes). I'm on
>>>>>>> parental
>>>>>>> leave until the end of November so I have limited bandwidth, but
>>>>>>> someone
>>>>>>> else should be able to pick it up.
>>>>>> I signed up to the FEniCS wiki so that I'll be able to add a short
>>>>>> guide.
>>>>>>
>>>>>> Alessio
>>>>>>
>>>>>>
>>>>>>> /Anders
>>>>>>>
>>>>>>>
>>>>>>>> Alessio
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> DOLFIN-dev mailing list
>>>>>>>> DOLFIN-dev at fenics.org
>>>>>>>> http://www.fenics.org/mailman/listinfo/dolfin-dev
>>>>>>
>>>>>> _______________________________________________
>>>>>> DOLFIN-dev mailing list
>>>>>> DOLFIN-dev at fenics.org
>>>>>> http://www.fenics.org/mailman/listinfo/dolfin-dev
>>>>>>
>>>>>
>>>>
>>>> _______________________________________________
>>>> DOLFIN-dev mailing list
>>>> DOLFIN-dev at fenics.org
>>>> http://www.fenics.org/mailman/listinfo/dolfin-dev
>>>>
>>>
>>
>>
>> _______________________________________________
>> DOLFIN-dev mailing list
>> DOLFIN-dev@xxxxxxxxxx
>> http://www.fenics.org/mailman/listinfo/dolfin-dev
>>
>
>