[Users] Carpet hg vs git

[Erik Schnetter]

Can you find out which call to MPI_Reduce causes this problem? Can you
then add print statements before the call to output the arguments? Can
you then look in the corresponding mpi.h file to see which datatype
etc. is used? If this is not a standard datatype, you could also
output the datatypes that Carpet creates (in dist.cc) to see which
datatype is actually causing this problem.

-erik

On Wed, Jul 27, 2011 at 4:05 PM, Ian Hinder <ian.hinder at aei.mpg.de> wrote:
>
> On 28 Jul 2011, at 01:04, Ian Hinder wrote:
>
>>
>> On 31 May 2011, at 03:57, Erik Schnetter wrote:
>>
>>> Luca
>>>
>>> Try finding out which datatype this is. I suspect this may be about a
>>> complex number. If so, I don't know why this is suddenly an error --
>>> as I mentioned before, the code to define the MPI datatype for a
>>> Cactus complex number didn't substantially change in the past years.
>>>
>>> The corresponding code is in dist.cc in the routine pseudoinit. Maybe
>>> we should be using the Fortran complex datatype for this, but Fortran
>>> support is not necessarily enabled in MPI (and we don't need it
>>> otherwise). Maybe we also need to define our own MPI operators for
>>> complex types.
>>
>> This is happening with the QuasiLocalMeasures test suite when run with the Mercurial version of Carpet on Damiana.  It doesn't happen with the Git version (same machine, compiler, optionlist, submit script, etc).  The only change to make this fail is to switch from the Git to the Mercurial version of Carpet.
>>
>> Luca's error message came just after a message from IsolatedHorizon (the thorn from which QuasiLocalMeasures was derived), so it's probably that which is triggering the error.
>
> The test output for one of the tests is at
>
>  http://damiana2.aei.mpg.de/~ianhin/testsuites/einsteintoolkit/einsteintoolkit.CarpetHG/2011-07-27-020826/QuasiLocalMeasures_qlm-bl.1.log
>
>
>>
>>>
>>> -erik
>>>
>>> On Sat, May 28, 2011 at 5:41 AM, Luca Baiotti <baiotti at ile.osaka-u.ac.jp> wrote:
>>>> On 28/5/11 1:38 PM, Erik Schnetter wrote:
>>>>>
>>>>> On Fri, May 27, 2011 at 9:03 PM, Luca Baiotti<baiotti at ile.osaka-u.ac.jp>
>>>>> wrote:
>>>>>>
>>>>>> Hi Erik,
>>>>>>
>>>>>> it seems to be Carpet or CarpetReduce:
>>>>>
>>>>> Luca
>>>>>
>>>>> Which MPI version is this? Is this a new MPI version? The
>>>>> corresponding code in Carpet hasn't changed in years, except for an
>>>>> added security check.
>>>>
>>>> It is openmpi 1.4.1, in particular, on damiana:
>>>>
>>>> openmpi/1.4.1-intel11.1.064
>>>>
>>>> Luca
>>>>
>>>
>>>
>>>
>>> --
>>> Erik Schnetter <schnetter at cct.lsu.edu>   http://www.cct.lsu.edu/~eschnett/
>>> _______________________________________________
>>> Users mailing list
>>> Users at einsteintoolkit.org
>>> http://lists.einsteintoolkit.org/mailman/listinfo/users
>>
>> --
>> Ian Hinder
>> ian.hinder at aei.mpg.de
>>
>
> --
> Ian Hinder
> ian.hinder at aei.mpg.de
>
>



-- 
Erik Schnetter <schnetter at cct.lsu.edu>   http://www.cct.lsu.edu/~eschnett/