[gpaw-users] Can anyone get mpi broadcast working?

[Ask Hjorth Larsen]

Hi

On Thu, 15 Dec 2011, Alex Eftimiades wrote:

> I thought it was working about a month ago. Now it does not. I have
> tried testing it with a simple broadcast, and the processes do not
> sync. There is no error, they just do not sync. I am beginning to
> suspect that there is something wrong in the source code since I
> updated between then and now. If not, please show me a working example
> (start to finish) that successfully broadcasts an ndarray with more
> than one element (it does seem to work for one element.)
> At the very least, the example provided in the documentation is
> incomplete. I tried copying and pasting it into a .py file and testing
> it. N and mydata need to be initialized with some values. So I
> initialized them with some values. Then I got the same problem as my
> tests--no errors, no syncing.
>
> For now, I may just have to write stuff to a file with one rank and
> load it with the others. However if this is a bug in the source code,
> I would imagine this could cause problems with parallel calculations.
>
> Thanks,
> Alex

How are you running the programme?  Through gpaw-python?  Broadcast is 
used in all parallel GPAW calculations, so the test suite should make 
hundreds of complaints if this function is not working.

Regards
Ask