Re: [PETSC #17275] Rank of the neighboring processors

[Barry Smith <bsmith@xxxxxxxxxxx>]

On Feb 6, 2008, at 6:38 AM, Tobias Kempe wrote:

> Hi Barry,
>
> i have to propose you an additional PETSc feature that could  
> interesting for
> some users. In our lagrangian particle tracking it is neccessary to  
> send the
> particle data to some specific processor - the new particle "master
> processor" which performs the evaluation of the equations of motion  
> for this
> particle.
>
> In general we can do this with the MPI_isend, MPI_irecv routines.
>
> In the PETSc environment this may be realized by an MPI_vector

   Do you mean a PETSc vector of type VECMPI, that is a parallel Vec?

> with different
> numer of entrys on each processor. The entrys of this MPI_vector may  
> be
> adressed by its index  and the rank of the processor where you wish  
> to set
> the value.

  Do you mean something like

    VecSetValues(vec,n,ranks,localindices,values)

where ranks[n], localindices[n], values[n]? Much like VecSetValues()  
with the "global indices"
replaced with ranks,localindices?

  Or do you want something more like a VecScatter operation where the  
"global indices"
are replaced with ranks,localindices

   Barry

> Perhaps it is possible to realize this feature in following petsc  
> releases.
>
> Best regards,
>
> Tobias
>
>
>
>
> Am Samstag 02 Februar 2008 schrieben Sie:
> >   Tobias,
> >
> >     I have added DAGetNeighbors().  You will need to use petsc-dev
> > see the developers
> > page link from www.mcs.anl.gov/petsc
> >
> >   Please let us know if you have any troubles.
> >
> >    Barry
> >
> > On Feb 1, 2008, at 3:00 AM, Tobias Kempe wrote:
> > > Hi PETSc team,
> > >
> > > we are using distributed arrays for the communication of the
> > > processors in our
> > > domain decomposition. Nomally the user dont has to worry obout the
> > > rank of
> > > the neighboring processors. Now we wont to compute large amounts of
> > > particles
> > > in viscous fluids. The problem is  when a particle leaves the  
> > > physical
> > > domain, an other processer becomes the "master" of this particle.
> > > With  the
> > > backround of this lagrangian particle tracking it becomes neccesarry
> > > to
> > > determine the ranks of the neighboring processors in order to send
> > > particle
> > > specific data to this processors.
> > >
> > > Is there some PETSc routine to dertermine the rank of the  
> > > neighboring
> > > processors in the east, west, front ..... and so on?
> > >
> > > Thank you very much.
> > >
> > > Best regards
> > >
> > > Tobias Kempe
> > >
> > >
> > >
> > >
> > > *****************************************************
> > > Dipl.-Ing. Tobias Kempe
> > > Technische Universität Dresden
> > > Institut für Strömungsmechanik
> > > Tel.:   0351-46336651
> > > Fax:   0351-46335246
> > > E-Mail: tobias.kempe@xxxxxxxxxxxxxxxxxxxx
> > > *****************************************************
>
>
>
> --
> *****************************************************
> Dipl.-Ing. Tobias Kempe
> Technische Universität Dresden
> Institut für Strömungsmechanik
> Tel.:   0351-46336651
> Fax:   0351-46335246
> E-Mail: tobias.kempe@xxxxxxxxxxxxxxxxxxxx
> *****************************************************