Wolfgang,

> > We are having trouble with the FE_DGQ<3> element (no problem occurs in
> > 2d).
> 
> I don't see this on my 48-core system. Is the only modification to step-12 
> the 
> use of 3d?

Yes.

> Guido changed something in the class/function you cite recently. Are you on 
> the latest svn revision?

We were able to run the same code (or rather the constructor of
FE_DGQ<3>(0), which is not working now) about a week ago. The problem
seems to be Guido's commit 22688 where he removed some locks and
introduced nested tasks (which seem to be the problem, not the removed
locks). If we undo the patch 22688, we get it running again. But the
tests (further down) still fail, and this does not appear to be a final
solution.

We have tested this on two systems and we got it only on one of them
(Opteron from 2007, 4 cores). It also did not depend on whether the
gcc-4.5 threads were used or the ones from boost.  

The error might be related to the problem that appears with the
testsuite on this system: There, the tests thread_validity_07, 08 and 09
fail, which are of similar structure as the part in
FETools::compute_embedding_matrices that fails. These tests have failed
already half a year ago, but we have never really bothered because all
user programs ran nicely.

=====Running======= thread_validity_07/exe
---------------------------------------------------------
In one of the sub-threads of this program, an exception
was thrown and not caught. Since exceptions do not
propagate to the main thread, the library has caught it.
The information carried by this exception is given below.

---------------------------------------------------------
Exception message: 
  
Exception type: 
  St12system_error
Aborting!
---------------------------------------------------------

=====Running======= thread_validity_08/exe
terminate called without an active exception
/bin/sh: line 1: 19666 Aborted                 ./thread_validity_08/exe
make: *** [thread_validity_08/output] Error 1

Best,
Katharina and Martin

_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii

Reply via email to