[Users] Output with carpet in more than 1 proc

[Ian Hinder]

On 24 Feb 2011, at 02:30, Erik Schnetter wrote:

> Francisco
> 
> Yes, the output format does (unfortunately) depend on the number of
> processes that you used. However, the content of the output should
> not!
> 
> There is a script "mergeCarpetIOASCII.pl" in thorn CarpetIOASCII that
> can help. (I'm not an expert here, I hope someone else can explain
> more.) For the HDF5 thorns, our visualisation tools should all be able
> to read the multiple files as if they were one.

Francisco,

Do you mean that the output format changes, or the output values?  If the former, then yes, this is very annoying, but it is by design.  If the latter, then this is something we have fought a lot with in the past.  It can occur due to changes in the order in which reduction operations occur caused by MPI and the lack of associativity of floating point arithmetic.  It can also happen due to the way that the compiler generates different code for handling different sizes of loop (e.g. vectorising in some cases, not in others, with the decision taken at runtime based on the size of the loop, which will depend on the number of processes).



> 
> -erik
> 
> On Wed, Feb 23, 2011 at 6:13 PM, Francisco S. Guzman
> <guzman at ifm.umich.mx> wrote:
>> 
>> Dear all,
>> 
>> I just coded a simple wavetoy and have found that the output when using
>> carpet (the number of time levels) depends on the number of
>> processors used for both h5 or ascii. Am I doing something clearly worng
>> and easily fixable?
>> 
>> Bests,
>> Francisco
>> --
>> Francisco S. Guzman Murillo.
>> Instituto de Fisica y Matematicas,
>> Universidad Michoacana de San Nicolas de Hidalgo
>> Phone: +52 (443) 322 3500 ext 4142
>> http://www.ifm.umich.mx/guzman/
>> _______________________________________________
>> Users mailing list
>> Users at einsteintoolkit.org
>> http://lists.einsteintoolkit.org/mailman/listinfo/users
>> 
> 
> 
> 
> -- 
> 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