[FLASH-USERS] GuardCell filling

[Seyit Hocuk]

Thought so.
Thanks for your reply.

Best, Seyit


On 02/19/2013 06:40 PM, Klaus Weide wrote:
> On Tue, 19 Feb 2013, Seyit Hocuk wrote:
>
>> I have a simple question on Grid_fillGuardCells.
>>
>> If I run a parallel job, for example on 2 processors, and I call the routine
>> "Grid_fillGuardCells(CENTER, ALLDIR, maskSize=NUNK_VARS, mask=gcMask,
>> minLayers=2, selectBlockType=LEAF)" in my block loop, what would happen if on
>> 1 processor I give a different mask (e.g. DENS_VAR) than on the other
>> processor (e.g. PRES_VAR). Or, DENS_VAR+PRES_VAR on one and only PRES_VAR on
>> the other. Would it still work?
> Hello Seyit,
> No, this would not work.
>
>> Also, what would happen if I call the routine on 1 proc but not on the other,
>> like "if (meshMe==0) call Grid_fillGuardCells(...)"? Would it freeze/crash?
> Again, this will not work, because under the hood Grid_fillGuardCells,
> and the underlying PARAMESH routines, do some communications in which all
> MPI tasks in MPI_COMM_WORLD (or more exactly, gr_meshComm) have to
> participate.
>
> Even if this appeared to work in some cases, it would not be doing the
> right thing.  All arguments to Grid_fillGuardCells have to agree, on all
> processes in gr_meshComm.
>
> Klaus


-- 
With kind regards,
Seyit


S. Hocuk
Postdoctoral Researcher
Kapteyn Astronomical Institute
P.O. box 9700
Groningen, the Netherlands