weird code in dbcsr

Noam Bernstein no... at bollweevil.gdbg.org
Fri Dec 30 17:01:54 UTC 2011


Never mind - I think I understand, more or less, now that I've thought about
what mpi_alltoall expects

    Noam

On Fri, Dec 30, 2011 at 11:28 AM, Noam Bernstein
<no... at bollweevil.gdbg.org> wrote:
> I'm trying to debug an intermittent crash, and in the process, I
> noticed the following
> in dbcsr_lib:
>
> dbcsr_lib/dbcsr_transformations.F, line 2586, calls mp_alltoall for
> send_count/recv_count
> passing count=2.  Not that send_count is allocated as (2,0:numnodes-1).
>
> mp_alltoall_i22 in dbcsr_lib/dbcsr_message_passing_i.F, line 257, passes
> count to mpi_alltoall.  I would have expected size(sb) and size(rb).
>
> Does this make sense, and if so, why?
>
> I'm using, I believe, the latest cvs version of the code.
>
>      thanks,
>      Noam



More information about the CP2K-user mailing list