[Dune] Identifying MPICH during configure

[Eike Mueller]

Hi Oliver,

thanks a lot, I just cloned the dune-common repository and switched to 
the releases/2.2 branch with "git checkout -b releases/2.2". I guess 
this is newer than the 2.2.1 tarball and contains further bugfixes? It 
already seems to contain the MPICH 3 fix already and compiles ok on ARCHER.

ARCHER ( http://www.archer.ac.uk/) has actually less cores than HECToR 
(~70,000) but they are Intel Ivybridge (instead of AMD Opterons) and 
roughly 2x faster (according to other's peoples experiences). On HECToR 
the DUNE multigrid is about a factor 2x slower than my handwritten 
Fortran code, so I want to check whether this difference disappears when 
I go to ARCHER. I also tried to play around with the compiler flags on 
HECToR, but there I never gain more than 5% (addmitedly I only changed 
the flags for my own DUNE multigrid module, and didn't recompile UG or 
DUNE-grid, but the code only has a 10%-20% overhead from iterating over 
the grid, so that should not make a big difference).

I'm hoping to still do a 80,000 core run on HECToR anyway before it is 
decommisioned in March.

Cheers,

Eike

On 24/01/14 08:25, Oliver Sander wrote:
> Am 24.01.2014 00:11, schrieb Eike Mueller:
>> Hi Steffen and Markus,
>>
>> thanks a lot, I looked at the mpi.h on my system and it indeed looks like it's MPICH 3.0 (there's a #define MPICH_VERSION "3.0.3"). Yes, I'm using version 2.2.1 of DUNE.
>>
>> As for UG, I was also using a slightly older version (patched with patch10-rc4), so I have now updated this with the latest patch as well, which fixes the problem there, I can configure, compile and install UG Grid with this.
>>
>> I also downloaded the 2.2.1 tarball of DUNE again and successfully applied the patch you describe for m4/mpi-config.m4. Unfortunately that does not seem to help, and I suspect this is because the tarball contains aclocal.m4, which does not contain the test_mpich3() marco. I don't know much about autotools, but when I try to delete aclocal.m4 and regenerate it with aclocal I run into a different problem:
>>
>> eike at eslogin004 $ aclocal
>> configure.ac:6: error: m4_defn: undefined macro: _m4_divert_diversion
>> configure.ac:6: the top level
>> autom4te: /usr/bin/m4 failed with exit status: 1
>> aclocal: autom4te failed with exit status: 1
> Hi Eike,
> quick solution:
> clone into dune-common, change to the 2.2 branch, apply the patch there and do a 'make dist'.
> That gives you a tarball with the patch.
> ideal solution: we do that for you, and do a 2.2.2 release of dune-common.
>
> PS: That Archer thing has how many processors?  What's your goal with UG?
>
> Cheers,
> Oliver [waiting for new bugreports :-) ]
>
>
>> Apart from this, just looking through the patch, running ./conftest in mpi_trybuild_run() might not work as the MPI compiler is configued to cross-compile for the compute nodes, so I'm not sure I can execute anything on the login nodes, but I will find out once the configure runs.
>>
>> Thanks,
>>
>> Eike
>>
>> On 23 Jan 2014, at 19:00, Steffen Müthing <steffen.muething at iwr.uni-heidelberg.de> wrote:
>>
>>> Am 23.01.2014 um 19:56 schrieb Markus Blatt <markus at dr-blatt.de>:
>>>
>>>> On Thu, Jan 23, 2014 at 07:49:20PM +0100, Steffen Müthing wrote:
>>>>> Hi Eike,
>>>>>
>>>>> what is the version of MPICH as reported by the command mpichversion or the module
>>>>> system? There was a problem with MPICH 3.x which has been fixed in the master branch
>>>>> of Dune (and I think also in the development version of UG), but I don't know whether those
>>>>> fixes have made it to the releases yet...
>>>>>
>>>> That one is in the release.
>>> Ah, that's good! :-)
>>>
>>> But if Eike is still on 2.2.1, he doesn't have that patch yet...
>>>
>>>>
>>>> -- 
>>>> Do you need more support with DUNE or HPC in general?
>>>>
>>>> Dr. Markus Blatt - HPC-Simulation-Software & Services http://www.dr-blatt.de
>>>> Hans-Bunte-Str. 8-10, 69123 Heidelberg, Germany
>>>> Tel.: +49 (0) 160 97590858  Fax: +49 (0)322 1108991658
>>>> _______________________________________________
>>>> Dune mailing list
>>>> Dune at dune-project.org
>>>> http://lists.dune-project.org/mailman/listinfo/dune
>>> _______________________________________________
>>> Dune mailing list
>>> Dune at dune-project.org
>>> http://lists.dune-project.org/mailman/listinfo/dune
>>
>> _______________________________________________
>> Dune mailing list
>> Dune at dune-project.org
>> http://lists.dune-project.org/mailman/listinfo/dune
>>
>
>
>
> _______________________________________________
> Dune mailing list
> Dune at dune-project.org
> http://lists.dune-project.org/mailman/listinfo/dune

-- 
Dr Eike Hermann Mueller
Research Officer (PostDoc)

Department of Mathematical Sciences
University of Bath
Bath BA2 7AY, United Kingdom

+44 1225 38 5633
e.mueller at bath.ac.uk
http://people.bath.ac.uk/em459/

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.dune-project.org/pipermail/dune/attachments/20140124/3a52c52c/attachment.htm>