dolfin team mailing list archive

[Johannes Ring <johannr@xxxxxxxxx>]

On Tue, Jun 22, 2010 at 11:04 AM, Johannes Ring <johannr@xxxxxxxxx> wrote:
> On Mon, Jun 21, 2010 at 11:52 PM, Anders Logg <logg@xxxxxxxxx> wrote:
>> On Mon, Jun 21, 2010 at 11:48:09PM +0200, Johannes Ring wrote:
>>> On Mon, Jun 21, 2010 at 10:52 PM, Anders Logg <logg@xxxxxxxxx> wrote:
>>> > On Mon, Jun 21, 2010 at 10:16:05PM +0200, Johannes Ring wrote:
>>> >> On Mon, Jun 21, 2010 at 9:12 PM, Anders Logg <logg@xxxxxxxxx> wrote:
>>> >> > On Mon, Jun 21, 2010 at 02:37:26PM +0200, Johannes Ring wrote:
>>> >> >> On Mon, Jun 21, 2010 at 1:53 PM, Anders Logg <logg@xxxxxxxxx> wrote:
>>> >> >> > On Wed, Jun 16, 2010 at 11:49:32AM +0200, Johannes Ring wrote:
>>> >> >> >> On Wed, Jun 16, 2010 at 11:27 AM, Anders Logg <logg@xxxxxxxxx> wrote:
>>> >> >> >> > When should the deadline be for making the new releases?
>>> >> >> >> >
>>> >> >> >> > Is it too later if we finish by June 23rd?
>>> >> >> >>
>>> >> >> >> Well, it might be too late for automatic imports from Debian to
>>> >> >> >> Ubuntu, but I can request manual imports later. It worked out fine the
>>> >> >> >> last time. So June 23rd sounds good for me.
>>> >> >> >
>>> >> >> > ok, let's aim for 23rd (Wednesday this week).
>>> >> >> >
>>> >> >> > Anything that we need to fix before the release?
>>> >> >> >
>>> >> >> > The buildbot fails with this message when running the extrapolation demo:
>>> >> >> >
>>> >> >> >  run-time error: auxlib::solve_od(): need LAPACK
>>> >> >>
>>> >> >> I have the same problem on my laptop.
>>> >> >>
>>> >> >> > Is there something that needs to be installed on the buildbot?
>>> >> >>
>>> >> >> I'm not sure whats wrong but the problem seems to be that I have built
>>> >> >> Armadillo with debugging turned on (-DCMAKE_BUILD_TYPE:STRING=Debug)
>>> >> >> on the buildbot (and on my laptop). When I run gdb on
>>> >> >> demo/adaptivity/extrapolation/cpp I get this:
>>> >> >>
>>> >> >> [...]
>>> >> >> #7  0xb76340b9 in dolfin::Extrapolation::compute_coefficients
>>> >> >> (coefficients=@0xbfe89fc0, v=@0xbfe8a258, V=@0xbfe8a3b8,
>>> >> >> W=@0xbfe8a338,
>>> >> >>     cell0=@0xbfe89f98, c0=@0xbfe89f58, dofs=@0xbfe89fd0, offset=@0xbfe89fcc)
>>> >> >>     at /home/johannr/src/armadillo-0.6.10/Build/local/include/armadillo_bits/debug.hpp:218
>>> >> >> #8  0xb7634e5b in dolfin::Extrapolation::extrapolate (w=@0xbfe8a178,
>>> >> >> v=@0xbfe8a258) at dolfin/adaptivity/Extrapolation.cpp:72
>>> >> >> #9  0xb740bd54 in dolfin::Function::extrapolate (this=0xbfe8a178,
>>> >> >> v=@0xbfe8a258) at dolfin/function/Function.cpp:330
>>> >> >> #10 0x0805c768 in main () at main.cpp:44
>>> >> >>
>>> >> >> Should I just rebuild Armadillo with debugging off on the buildbots?
>>> >> >>
>>> >> >> Johannes
>>> >> >
>>> >> > Could you add some debugging to Extrapolation.cpp to see where things
>>> >> > go wrong? Find the line that says
>>> >> >
>>> >> >  arma::Col<double> x = arma::solve(A, b);
>>> >> >
>>> >> > and print something before and after. It would be interesting to see
>>> >> > if
>>> >> >
>>> >> >  1. This is where it breaks
>>> >> >  2. It breaks already the first time or just for particular input data
>>> >>
>>> >> Yes, this is where it breaks and it breaks the first time. Anything
>>> >> else I should try?
>>> >>
>>> >> Johannes
>>> >
>>> > Can you check if you are able to run call arma::solve(A, b) at all?
>>> >
>>> >  arma::mat A(M, N);
>>> >  arma::vec b(M);
>>> >
>>> >  then set some bogus values in A and b
>>> >
>>> >  arma::Col<double> x = arma::solve(A, b);
>>> >
>>> > By the way, why is the output arma::Col and not arma::vec? Garth?
>>> >
>>> > If this works, then try to print the size of A and b in
>>> > Extrapolation.cpp and try the same size in your test program.
>>>
>>> It doesn't work. I tried this test program
>>>
>>>   arma::mat A = arma::rand(4,4);
>>>   arma::vec b = arma::rand(4);
>>>   arma::vec x = arma::solve(A, b);
>>>
>>> and I got the same problem.
>>>
>>> Johannes
>>
>> Works for me. I've tried the following test program:
>>
>> #include <armadillo>
>>
>> int main()
>> {
>>  arma::mat A = arma::rand(4,4);
>>  arma::vec b = arma::rand(4);
>>  arma::vec x = arma::solve(A, b);
>>
>>  return 0;
>> }
>
> Yes, that's what I did. I also tried to build Armadillo (several
> different versions) without debugging and I still got the same runtime
> error.
>
>> Compiled with
>>
>> g++ main.cpp -llapack
>>
>> Runs with with and without gdb. Also works fine with valgrind.
>>
>> Anyone else has this problem? I'm using the Armadillo Ubuntu package.
>
> Maybe you can try to build Armadillo? Here are the steps:
>
>  wget http://downloads.sourceforge.net/project/arma/armadillo-0.9.10.tar.gz
>  tar zxf armadillo-0.9.10.tar.gz
>  cd armadillo-0.9.10/
>  mkdir build
>  cd build
>  cmake -DCMAKE_INSTALL_PREFIX:PATH=$PWD/local ..
>  make && make install
>  g++ -I$PWD/local/include -L$PWD/local/lib64 -larmadillo /path/to/test.cpp
>  LD_LIBRARY_PATH=$PWD/local/lib64 ./a.out

I found out that this out-of-source build does not work correctly with
Armadillo. These steps works fine on all my setups:

wget http://downloads.sourceforge.net/project/arma/armadillo-0.9.10.tar.gz
tar zxf armadillo-0.9.10.tar.gz
cd armadillo-0.9.10/
cmake -DCMAKE_INSTALL_PREFIX:PATH=$PWD/local \
  -D LIB_INSTALL_DIR:STRING=lib .
make && make install
g++ -I$PWD/local/include -L$PWD/local/lib -larmadillo /tmp/test.cpp
LD_LIBRARY_PATH=$PWD/local/lib ./a.out

The buildbot is green now on jaunty-amd64 and will hopefully also be
green on hardy-i386. The mac-osx buildslave will not be green because
of some MPI problems (I will try to fix this after the summer).

Johannes

>> Shouldn't that be used on the buildbot?
>
> Yes, but Armadillo wasn't added to Ubuntu before Karmic and the Ubuntu
> buildbots are older than that. My plan was to upgrade them but the xen
> server running on the host has problems (according to the sysadmins at
> Simla) when upgrading from Jaunty to Karmic. The plan is therefore to
> upgrade the host and then the buildbot slaves, but I'm not sure when
> it will happen.
>
> Johannes
>