New Binary Package for MAC

[Luca Heltai]

Dear All,

since some of you have been experiencing problems with the last binary package for MAC, I created a new one here:

https://github.com/luca-heltai/dealii/releases/tag/v8.2pre_v3

where I specified all compilers with absolute path. This should be compatible also with machines where macports compilers are also installed.

Best,

Luca.

--
Luca Heltai <luca....@gmail.com>
http://people.sissa.it/~heltai/
Scuola Internazionale Superiore di Studi Avanzati
Phone: +39 040 3787 449, Office: 622
--
There are no answers, only cross references

[Timo Heister]

Dear Luca,

some mac ASPECT users have problems with the new package, see for
example (.txt attachment linked at the bottom):
https://geodynamics.org/pipermail/aspect-devel/2014-October/000706.html

Also (this looks like the Trilinos matrix assemble bug):
https://geodynamics.org/pipermail/aspect-devel/2014-October/000705.html


Do you have an idea what is going on?

> --
> The deal.II project is located at http://www.dealii.org/
> For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en
> ---
> You received this message because you are subscribed to the Google Groups "deal.II User Group" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to dealii+un...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.



--
Timo Heister
http://www.math.clemson.edu/~heister/

[Luca Heltai]

It seems from aspect mailing list that the problem is related to a buggy Trilinos version. Can you tell me what's the last Trilinos version known to work without problems?

I'll repackage everything with the specified version, and ship a v8.2pre_v4 asap.

L.

--
Luca Heltai <luca....@gmail.com>
http://people.sissa.it/~heltai/
Scuola Internazionale Superiore di Studi Avanzati
Phone: +39 040 3787 449, Office: 622
--
There are no answers, only cross references

[Timo Heister]

I think it is only 11.10 that is broken, so 11.8 should work.

[Martin Kronbichler]

Timo,

> I think it is only 11.10 that is broken, so 11.8 should work.

Do you have a version where you can reproduce the bug listed on the
Aspect list? My suspicion is not that 11.10 is broken but that we try to
use a feature in 11.10 that we do not in 11.8: In line 366 of
trilinos_sparsity_pattern.cc we enable a flag 'nonlocalGraph' of the
Trilinos sparsity pattern which propagates to the matrix. I thought I
had worked around the problems in late September but apparently it was
not enough on all platforms. So I'd like you to test what happens if you
set the check to
#if DEAL_II_TRILINOS_VERSION_GTE(11,13,0)
(it appears that the issue will not be fixed for 11.12 either). Even
though this definitely is a bug in Trilinos, I don't like black-listing
every other release but we should work around them.

Best,
Martin

[Timo Heister]

> Do you have a version where you can reproduce the bug listed on the
> Aspect list?

No, I don't even have a mac. If you think this will also happen on
linux, I will give it a try.

[Martin Kronbichler]

Timo,

> > Do you have a version where you can reproduce the bug listed on the
> > Aspect list?
>
> No, I don't even have a mac. If you think this will also happen on
> linux, I will give it a try.

The bug described on the aspect list does indeed look like it should
also show up in Linux with Trilinos 11.10 and deal.II master. You should
be able to check if it works there or not (or let me know which case I
should run).

Best,
Martin

[Matthias Maier]

I've upgraded simserv04 to trilinos 11.10.2 yesterday. Here is a result:

http://cdash.kyomu.43-1.org/buildSummary.php?buildid=447


What exactly is this bug?

Best,
Matthias

[Martin Kronbichler]

> I've upgraded simserv04 to trilinos 11.10.2 yesterday. Here is a result:
>
> http://cdash.kyomu.43-1.org/buildSummary.php?buildid=447
>
>
> What exactly is this bug?

tests/trilinos_assemble_matrix_parallel_07 fails which is supposed to
test it. So Timo, it will very likely show in aspect if you initialize
the matrix via


I will change the trilinos sparsity pattern now
>
> Best,
> Matthias
>

[Martin Kronbichler]

> I've upgraded simserv04 to trilinos 11.10.2 yesterday. Here is a result:
>
> http://cdash.kyomu.43-1.org/buildSummary.php?buildid=447
>
>
> What exactly is this bug?

tests/trilinos_assemble_matrix_parallel_07 fails which is supposed to

test whether we can write into the Trilinos matrix in parallel. This
does not seem to be the case as the test shows. I'm a bit unsure why
only the 07 test fails - I would also have expected some of the other
tests to fail because what is supposed to be the 'reference' matrix
should also be wrong - it might be because we call 'compress' twice in
the test and do not do it in the other tests.

So Timo, it will very likely show in aspect if you initialize the

sparsity pattern of the Trilinos matrix by
csp.reinit(locally_owned, locally_owned, communicator);

I will change the define in trilinos sparsity pattern now because it
definitely leads to wrong results that we can see in our test suite. A
pity that I always work against a local Trilinos version where I fixed
the issue in the Trilinos source code and hence did not see the problem.

Best,
Martin

[Wolfgang Bangerth]

> tests/trilinos_assemble_matrix_parallel_07 fails which is supposed to
> test whether we can write into the Trilinos matrix in parallel. This
> does not seem to be the case as the test shows. I'm a bit unsure why
> only the 07 test fails - I would also have expected some of the other
> tests to fail because what is supposed to be the 'reference' matrix
> should also be wrong - it might be because we call 'compress' twice in
> the test and do not do it in the other tests.
>
> So Timo, it will very likely show in aspect if you initialize the
> sparsity pattern of the Trilinos matrix by
> csp.reinit(locally_owned, locally_owned, communicator);
>
> I will change the define in trilinos sparsity pattern now because it
> definitely leads to wrong results that we can see in our test suite. A
> pity that I always work against a local Trilinos version where I fixed
> the issue in the Trilinos source code and hence did not see the problem.

Bummer. Are you saying that you've got it fixed locally for yourself but that
they didn't include your patch into 11.10? Did you upload it to the trilinos
bug report?

Thanks
Wolfgang


--
------------------------------------------------------------------------
Wolfgang Bangerth email: bang...@math.tamu.edu
www: http://www.math.tamu.edu/~bangerth/

[Martin Kronbichler]

> Bummer. Are you saying that you've got it fixed locally for yourself but that
> they didn't include your patch into 11.10? Did you upload it to the trilinos
> bug report?

I submitted a bug report about a month ago which fixed the problems I
observed, but there might be more and I don't know why it got not fixed
yet. My report might have been too close to the 11.12 release...

In any case, when we also pass in the locally relevant dofs to the
reinit, we get the thread-safe path also now, so the issue is not a
showstopper.

Best,
Martin