[Mpi-forum] Communicator info arguments

[Jeff Hammond]

In 6.4.4, I find this:

"Info hints are not propagated by MPI from one communicator to another
except when the communicator is duplicated using MPI_COMM_DUP or
MPI_COMM_IDUP. In this case, all hints associated with the original
communicator are also applied to the duplicated communicator."

I think this resolves any ambiguity.

Jeff

On Fri, Dec 7, 2012 at 4:13 PM, Pavan Balaji <balaji at mcs.anl.gov> wrote:
>
> Right.  I'm trying to understand what the intention was.
>
> The intention stated at the Forum might not be accurate.
>
> Either way, it needs to be stated in the standard, what the application
> should expect.
>
>  -- Pavan
>
> On 12/07/2012 04:11 PM US Central Time, Jeff Hammond wrote:
>> Okay, sorry, I didn't understand your question properly.
>>
>> I think your example of intercomm_merge indicates that propagating
>> info with functions other than dup is problematic and probably not
>> easily specified.  The safest thing to do is to say that they do not
>> propagate.
>>
>> Jeff
>>
>> On Fri, Dec 7, 2012 at 4:04 PM, Pavan Balaji <balaji at mcs.anl.gov> wrote:
>>>
>>> My question was for the other communicator creation functions.
>>> Specifically, Comm_split in my query below.
>>>
>>>  -- Pavan
>>>
>>> On 12/07/2012 04:01 PM US Central Time, Jeff Hammond wrote:
>>>> >From MPI-3 6.4.2 (MPI_COMM_DUP):
>>>>
>>>> "MPI_COMM_DUP duplicates the existing communicator comm with the
>>>> associated key values, topology information and info hints."
>>>>
>>>> "MPI_COMM_DUP_WITH_INFO behaves exactly as MPI_COMM_DUP except that
>>>> the info hints associated with the communicator comm are not
>>>> duplicated in newcomm.  The hints provided by the arguments info are
>>>> associated with the output communicator newcomm instead."
>>>>
>>>> Jeff
>>>>
>>>> On Fri, Dec 7, 2012 at 3:56 PM, Pavan Balaji <balaji at mcs.anl.gov> wrote:
>>>>>
>>>>> Apologies if this query needs to go a specific working group.  I wasn't
>>>>> sure which one, so I'm sending it to the overall mailing list.
>>>>>
>>>>> If a communicator is created using MPI_COMM_DUP_WITH_INFO, is the info
>>>>> argument "sticky" to all child communicators?  That is, if I split the
>>>>> output communicator further, will the info argument propagate?
>>>>>
>>>>> Adam said that it would, at the Forum, but that doesn't seem accurate,
>>>>> since that'll get weird if I do an Intercomm_merge with two
>>>>> communicators, one of which has an info argument and the second does not.
>>>>>
>>>>> I couldn't find text saying one way or another in the document.
>>>>>
>>>>>  -- Pavan
>>>>>
>>>>> --
>>>>> Pavan Balaji
>>>>> http://www.mcs.anl.gov/~balaji
>>>>> _______________________________________________
>>>>> mpi-forum mailing list
>>>>> mpi-forum at lists.mpi-forum.org
>>>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum
>>>>
>>>>
>>>>
>>>
>>> --
>>> Pavan Balaji
>>> http://www.mcs.anl.gov/~balaji
>>
>>
>>
>
> --
> Pavan Balaji
> http://www.mcs.anl.gov/~balaji



-- 
Jeff Hammond
Argonne Leadership Computing Facility
University of Chicago Computation Institute
jhammond at alcf.anl.gov / (630) 252-5381
http://www.linkedin.com/in/jeffhammond
https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond