(j3.2006) (SC22WG5.5138) image selectors

[Tom Clune]

On Dec 9, 2013, at 6:52 PM, Van Snyder <Van.Snyder at jpl.nasa.gov> wrote:

> On Mon, 2013-12-09 at 17:18 -0600, Bill Long wrote:
>> 
>> On 12/9/13 5:05 PM, Van Snyder wrote:
>>>> OK.  Reinhold's revised ballot reworded this idea in terms of the image
>>>>> index in the initial team rather than physical processors.  That is
>>>>> arguably better terminology to use.  The image's image index in the
>>>>> initial team never changes throughout the program execution.
>> 
>>> I assume this refers to Reinhold's message of 2 December.  That
>>> message's attachment did not include any comments concerning 5.1.  The
>>> problem is that "image indices are relative to a specified team" at
>>> [9:5-6] does not give any information concerning the correspondence
>>> between coindices in parent teams and subteams, nor does "cosubscripts
>> 
>> See [11:21-22].  The program can explicitly specify the new image index. 
>> Otherwise it is processor dependent.
> 
> I think that means that in most applications one must specify NEW_INDEX.
> Why make it optional?
> 
> Instead of "assigned by the processor" the term "processor dependent"
> should be used.
> 
> Better yet would be to specify the mapping from parent team to subteam.
> If NEW_INDEX is not specified, is there really a difficulty with
> specifying, for example, that the image indices for the subteam are in
> the same order as the image indices in the parent team, so that image
> index 1 for the subteam applies to the image with the smallest image
> index in the parent team that becomes part of the subteam, etc.?

I'll second that.   ince many applications will be ported from MPI, the principle of "least surprise" should be invoked here.  Unless there is a reason to the contrary, the MPI convention for subcommunicator ranks should apply to subteam image indices.   MPI allows redefining ranks as well, but I've only used that once (maybe) in 20+ years of coding in MPI.

- Tom



> 
>>> are interpreted as if the current team were the team specified by
>>> <team-variable>" at [11:4].  Without standardizing this, indexing with
>>> respect to ancestor teams is not useful.  I tried in vain to find this
>>> mapping in 5.3 -- 5.5.  The addition of DISTANCE to THIS_IMAGE doesn't
>>> seem to do the job.
> 
> 
> _______________________________________________
> J3 mailing list
> J3 at mailman.j3-fortran.org
> http://mailman.j3-fortran.org/mailman/listinfo/j3

Thomas Clune, Ph. D. 					<Thomas.L.Clune at nasa.gov>
Chief, Software Systems Support Office		Code 610.3
NASA GSFC								301-286-4635
MS 610.8 B33-C128						<http://ssso.gsfc.nasa.gov>
Greenbelt, MD 20771





-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.j3-fortran.org/pipermail/j3/attachments/20131209/8d13d5f6/attachment-0001.html