MPI_GATHER can be used to put data in this order, but you need to think
about your data layout in memory, and structure your use datatypes and
MPI_GATHER accordingly.
Keep in mind that the datatype that you gather *from* does not need to
be the same datatype that you gather *to*. Specifically, the datatypes
provided on the root and non-root processes do not need to be the same.
They must be equivalent -- essentially meaning that the resulting
number of bytes and basic data elements are the same, but the layout in
memory could be different.
I realize that I'm not giving you much of an answer :-(, but datatypes
are a rather complex issue; I would strongly suggest spending a little
time with an MPI book or a tutorial to get at least the basics of
complex datatypes. They can give much more insight and detail than I
can in an e-mail.
Hope that helps!
On Apr 15, 2005, at 7:05 PM, Khaled Al Salem wrote:
> Hello,
> I wrote an f90 mpi code to solve a 3D problem. The topology is
> obtained using mpi_cart_creat. At the end of the calculations I wish
> to get the results gathered from all processors into one big 3
> dimensional array that can be directly visualized without going
> through post-processing. The problem is that mpi_gather doesn't
> gather the results from the different proccessors as blocks.
> for example, in 2D, if proc. 1 has the following
>
> 1 1 1
> 1 1 1
> 1 1 1
>
> and proc. 2
>
> 2 2 2
> 2 2 2
> 2 2 2
>
> then mpi_gather results in the following
>
> 1 1 1 1 1 1
> 1 1 1 2 2 2
> 2 2 2 2 2 2
>
> the results that I'm after would be
>
> 1 1 1 2 2 2
> 1 1 1 2 2 2
> 1 1 1 2 2 2
>
> is there an easy way to do this?
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
--
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
|
