[mpich-discuss] Poor scaling of MPI_WIN_CREATE?

[Jeff Hammond]

I asked for CrayMPI source already and the answer was and always will
be "no".  I can elaborate offline.

I've discussed the implementation enough with Cray to know that DMAPP
is not the basis for their RMA implementation, or at least was not 6
months ago.

Jeff

On Wed, May 30, 2012 at 11:16 AM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
> On Wed, May 30, 2012 at 11:14 AM, Timothy Stitt <Timothy.Stitt.9 at nd.edu>
> wrote:
>>
>> This is a good point Jed. Can I be sure that the current xt-mpich2 library
>> on the Cray systems doesn't exploit the DMAPP routines already...before I
>> invest time in writing something myself?
>
>
> Well, if they were good citizens and released their code (even IBM does
> this), it would be trivial to find out. As it is, you can either ask them
> (and take their word for it) or read the disassembly.
>
>>
>>
>> Isn't the whole point of a "vendor optimized MPI" that they would have
>> done this? Is there a semantic reason why MPI_Win_create() cannot be
>> implemented in this fast way using DMAPP?
>>
>>
>> Tim Stitt PhD (User Support Manager).
>>
>> Center for Research Computing | University of Notre Dame |
>> P.O. Box 539, Notre Dame, IN 46556 | Phone:  574-631-5287 | Email:
>> tstitt at nd.edu
>>
>>
>> _______________________________________________
>> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
>> To manage subscription options or unsubscribe:
>> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>>
>
>
> _______________________________________________
> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
> To manage subscription options or unsubscribe:
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>



-- 
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