[MITgcm-devel] exch2 problem?

[Martin Losch]

That sounds good to me. Will you do it, or should I do in config_check with an appropriate message?

Martin

On Mar 24, 2012, at 2:46 PM, Jean-Michel Campin wrote:

> Hi Martin,
> 
> In general, there is a potential problem when Olx > sNx, or Oly > sNy,
> i.e., not enough value in the tile interior to fill-in the overlap.
> And this is the reason why we have this special piece of code to
> deal with cases Nx = 1 or Ny = 1 (eesupp/src/exch1_rx.template, after
> line 240), but this special piece of code is not there in pkg/exch2.
> May be this could be the reason ?
> 
> I propose to add a stop for these 2 cases: Olx > sNx, or Oly > sNy,
> except if EXCH1 and Nx=1 or Ny=1.
> 
> Cheers,
> Jean-Michel
> 
> On Sat, Mar 24, 2012 at 08:28:15AM +0100, Martin Losch wrote:
>> Hi there,
>> 
>> Yossi is running a zonally averaged model with Nx=1, Ny = 84, Nr=120 on a lat-lon grid. I attached a simple configuration which starts from uniform salinity/temperature and has some heat flux to make the model move.
>> 
>> We find that it works with one tile, but not with more than one (the attached SIZE.h uses 2 tiles on 1processor, but 2processore with 1 tile each gives the same). It explodes once the cg2d solver has something to do (basically cg2d no longer converges). Yossi finds that it also works with more than one tile, if he increases: sNx=2.
>> 
>> The problem goes away when I recompile without exch2, so it seems to be related to exch2. While I don't think that exch2 has to work with such a configuation (and exch2 is only in there because Yossi went from a cs32 configuration to this lon-lat configuration), I do think it would be good, if we could either fix the problem, or add a stop when (sNx=1||sNy=1)&&exch2. What do you think? [Maybe I overlooked something and there is problem with the configuration, but not with exch2?]
>> 
>> Martin
>> 
>> 
> 
> 
>> _______________________________________________
>> MITgcm-devel mailing list
>> MITgcm-devel at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
> 
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel

Martin Losch
Martin.Losch at awi.de