[Q-e-developers] Bug Affecting Hybrid Calculations

[Paolo Giannozzi]

Hi Taylor (and CC: to q-e-developers)

please try the attached patch (I cannot make any serious test right now).
The initialization of the so-called "custom grid" was slightly different on
different pools, in addition to being exceedingly confused and confusing.

Paolo

On Sun, Aug 13, 2017 at 5:32 AM, Taylor Barnes <tbarnes at lbl.gov> wrote:

> Hi Paolo,
>
>    I've recently encountered a somewhat strange bug that crashes certain
> hybrid calculations.  I had originally thought that it was a consequence of
> my changes to exx.f90, but I am able to reproduce the bug in the public
> release of 6.0, which does not include my changes.  The bug does not seem
> to exist in QE 5.4.0.
>
>    When running certain hybrid calculations with pools, the code crashes
> near the end of subroutine exxinit, at the point when exxbuff is summed
> across pools (at line 846 of exx.f90 in QE 6.0).  The error message is
> along the lines of:
>
> Message from rank 7 and tag 14 truncated; 358400 bytes received but buffer
> size is 302976
>
>    After some more investigating, I have found that this happens because
> nrxxs is somehow different for each pool, which of course means that the
> size of exxbuff different for each pool, and inter-pool communication fails
> as a result.  As best as I can tell, this issue was introduced by some of
> the changes that were made to FFTXlib between QE 5.4 and QE 6.0.  I'm
> having a hard time narrowing things down much further than that, partly due
> to my limited understanding of those changes.
>
>    I have attached an input file for a small calculation that is affected
> by the bug.  I ran on a total of 8 MPI tasks and used 8 pools.
>
>    Do you have any insight into what might be happening here?
>
> Thanks,
> Taylor
>
>
>


-- 
Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20170814/b181ebe7/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: exx.patch
Type: text/x-patch
Size: 9329 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20170814/b181ebe7/attachment.bin>