[MITgcm-support] No. of processes not equal to nPx*nPy

[Jean-Michel Campin]

Hi Gus,

On Tue, Apr 21, 2009 at 07:52:11PM -0400, Gus Correa wrote:
> Hi Jean-Michel
>
> I keep a bunch of SIZE_Np.h_mpi
> (where N is the number of processors to use).
> My SIZE.h is just a soft link to that particular number
> of processors I want to use.
> This is very convenient when you have to accommodate with
> other user jobs in a cluster with a batch queue,
> shrinking or expanding the number of requested processes
> depending on their availability and queue status.
>
> The new genmake2 scheme will break this, right?

It will not break. you will see (ls -l SIZE*) links:
 SIZE_Np.h -> ../code/SIZE_Np.h_mpi
and also:
 SIZE.h -> ../code/SIZE.h
in your build dir, but this will have no effect since 
"SIZE_Np.h" is never included in any source file.
The only one to avoid is "SIZE.h_mpi".

> What is the upside of the new genmake2 scheme?
> Can it be reverted to the old one?

We will consider the option of going back to the previous (simpler)
behavior.

Thanks,
Jean-Michel

> Of course I can rename the files SIZE_Np.h_mpi with a
> different suffix.
> However, if the decision to use different include files
> is based on slightly different file names,
> and is hidden inside genmake2,
> this may trick users, as it may have done to Nico.
> Including a single SIZE.h with the standard file name
> avoids confusion.
>
> Just a suggestion.
>
> Thank you,
> Gus Correa
> ---------------------------------------------------------------------
> Gustavo Correa
> Lamont-Doherty Earth Observatory - Columbia University
> Palisades, NY, 10964-8000 - USA
> ---------------------------------------------------------------------
>
>
> Jean-Michel Campin wrote:
>> Hi Nico,
>>
>> Can you check which "SIZE.h" is in your directory where you build the 
>> executable ?
>>
>> I realized that I made a modification to genmake2 (and still not sure 
>> if this was a good idea) which could explain
>> the changes you see now:
>> with the option "-mpi", genmake2 is now linking *_mpi files from all
>> our "mods-dir" and then renames the link without the suffix "_mpi".
>> If you keep a "SIZE.h_mpi" in your mods-dir, it was ignored before
>> but is now linked before "SIZE.h". You can just remove "SIZE.h_mpi"
>> and everything will work as they used to.
>>
>> Thanks,
>> Jean-Michel
>>
>> On Tue, Apr 21, 2009 at 12:51:50PM -0400, Nico Wienders wrote:
>>>
>>>
>>>
>>>
>>> MITgcmers,
>>>
>>> I got a funny one this morning:
>>>
>>> (PID.TID 0003.0001) *** ERROR *** S/R EEBOOT_MINIMAL: No. of 
>>> processes  not equal to nPx*nPy    12     2
>>>
>>> In this message i believe the first column (12)  indicate the number 
>>> of procs i request in my mpi command
>>> The second column (2) is on how many processors the code recognizes 
>>> it  should run. Correct?
>>>
>>> But my SIZE.h is:
>>>
>>>     &           sNx =  150,
>>>     &           sNy =  150,
>>>     &           OLx =   3,
>>>     &           OLy =   3,
>>>     &           nSx =   1,
>>>     &           nSy =   1,
>>>     &           nPx =   3,
>>>     &           nPy =   4,
>>>     &           Nx  = sNx*nSx*nPx,
>>>     &           Ny  = sNy*nSy*nPy,
>>>     &           Nr  =   140)
>>>
>>>
>>>
>>> I do not understand. These configurations were all working fine in 
>>> the  past. Thank you.
>>> _______________________________________________
>>> MITgcm-support mailing list
>>> MITgcm-support at mitgcm.org
>>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>> _______________________________________________
>> MITgcm-support mailing list
>> MITgcm-support at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support