Re: [Boost-mpi] cartesian ecomposion

Subject: Re: [Boost-mpi] cartesian ecomposion
From: Alain Miniussi (alain.miniussi_at_[hidden])
Date: 2014-09-22 12:10:31

There is a branch named cartesian with a tentative implementation/test. 
Since that does not fall in the bugfix category, I'd like to have some 
opinion about it before pushing on the develop branch.

Any comment welcomed but there are two point for which I would 
specifically appreciate some input:

   - As for the topology description, the C api does it through two 
separate int arrays, one for the dimensions and one for the periodicity 
(which must be of the same size obviously). I choose to do it through 
one single array of pod struct { int size; bool periodic;} since I 
though it made to association betwen size and periodicity cleared and 
reduced the risk of a array size mismatch. But it does make two more 
classes.
  - One could imagine to add a templatized version that would fix the 
number of dimension at compile time, but I could not convince myself of 
the relevance of such a class. In the use cases I could think of, it 
wasn't clear that the benefit out-weighted the add complexity.

Nite that there is not mapping for shift yet, since it seems to be used 
with sendrecv for which there is no mapping yet.

Enjoy (well sort of)

Alain

On 10/09/2014 13:53, Alain Miniussi wrote:
> On 03/09/2014 13:27, Hal Finkel wrote:
>> Alain,
>>
>> I believe it would be useful to support the Cartesian communicators, 
>> it is just that no one has done the development work. Please feel 
>> free to rectify that shortcoming :-)
>
> Ok, I'll look into it.
> Right now I have a few issues that shows up with Intel's MPI that 
> makes the non-blocking test fail (hang actually).
> I made 3 pull requests related to those.
> I'll try starting working on cartesian once they're fixed since I'd 
> like to have all the test passing on my development machine before that.
>
> Alain
>
>>
>>   -Hal
>>
>> ----- Original Message -----
>>> From: "Alain Miniussi" <alain.miniussi_at_oca.eu>
>>> To: boost-mpi_at_[hidden]
>>> Sent: Wednesday, September 3, 2014 3:44:03 AM
>>> Subject: [Boost-mpi] cartesian ecomposion
>>>
>>> Hi
>>>
>>> The Cartesian decomposition part of the MPI spec does not seems to be
>>> present in boost mpi.
>>> Is it just due to a lack of resources or is it useless for some
>>> reason ?
>>>
>>> I'd like to look into it but would like to make sure there is a
>>> point.
>>>
>>> Best regards
>>>
>>> Alain
>>>
>>> _______________________________________________
>>> Boost-mpi mailing list
>>> Boost-mpi_at_[hidden]
>>> http://lists.boost.org/mailman/listinfo.cgi/boost-mpi
>>>
>
>

-- 
---
Alain