[gpaw-users] Running GPAW tests during build
Jens Jørgen Mortensen
jensj at fysik.dtu.dk
Mon Jun 20 10:18:58 CEST 2016
On 06/18/2016 08:37 PM, Marcin Dulak wrote:
> Hi,
>
> on Fedora I noticed that "parallel no" is reported by the
> tests, what can be causing this?
> Note that "scalapack yes" and parallel/submatrix_redist.py runs.
It's because gpaw-python is not in $PATH.
> mpiexec -np 4 build_openmpi/bin.linux-x86_64-2.7/gpaw-python
> /home/vagrant/rpmbuild/BUILD/gpaw-1.0.0/python2/tools/gpaw-test
> --range=linalg/gemm_complex.py,fileio/restart.py --debug
> which: no gpaw-python in
> (/home/vagrant/rpmbuild/BUILD/gpaw-1.0.0/python2/tools:/home/vagrant/rpmbuild/BUILD/gpaw-1.0.0/python3/tools:/usr/lib64/openmpi/bin:/home/vagrant/rpmbuild/B
> UILD/gpaw-1.0.0/python2/tools:/home/vagrant/rpmbuild/BUILD/gpaw-1.0.0/python3/tools:/usr/local/bin:/usr/bin:/usr/local/sbin:/usr/sbin:/home/vagrant/.local/bin:/home/vagrant/bin)
> python-2.7.11
> /home/vagrant/rpmbuild/BUILD/gpaw-1.0.0/python2/build_openmpi/bin.linux-x86_64-2.7/gpaw-python
> gpaw-1.0.0
> /home/vagrant/rpmbuild/BUILD/gpaw-1.0.0/python2/build/lib.linux-x86_64-2.7/gpaw/
> ase-3.11.0 /usr/lib/python2.7/site-packages/ase/
> numpy-1.9.2 /usr/lib64/python2.7/site-packages/numpy/
> scipy-0.14.1 /usr/lib64/python2.7/site-packages/scipy/
> _gpaw built-in
> parallel no
> FFTW no
> scalapack yes
> libvdwxc no
> PAW-datasets /usr/local/share/gpaw-setups:
> /usr/share/gpaw-setups
> Running tests in /tmp/gpaw-test-QfZtl2
> Jobs: 1, Cores: 4, debug-mode: True
> =============================================================================
> ...
> fileio/hdf5_noncontiguous.py 0.065 OK
> eigen/cg2.py 0.059 OK
> fd_ops/laplace.py 0.000 SKIPPED
> ...
> parallel/submatrix_redist.py 0.045 OK
>
>
> Two more questions:
>
> 1. is hdf5 expected to work with python3?
No, it has not been ported to Python 3. It's not high priority to port
it - we could simply switch to h5py instead.
>
> 2. python 3 does not have libpython3.4.so <http://libpython3.4.so>,
> only libpython3.4m.so <http://libpython3.4m.so>. Some systems
> (Debian/Ubuntu) provide a symbolic link to libpython3.4m.so
> <http://libpython3.4m.so>, but that's not the case of Fedora.
> Can you make a survey on the gpaw-users if all python versions provide
> get_config_vars()['BLDLIBRARY'] so it can be used in config.py instead
> of assuming '-lpython%s' % cfgDict['VERSION']?
>
> $ python3 -c "from distutils.sysconfig import get_config_vars;
> print(get_config_vars()['BLDLIBRARY'])"
> -L. -lpython3.4m
> $ python -c "from distutils.sysconfig import get_config_vars;
> print(get_config_vars()['BLDLIBRARY'])"
> -L. -lpython2.7
It works OK on the computers I have access to. I would say it's OK to
do as you suggest. If you already tried this then please send me a
patch or create a merge request.
Thanks!
Jens Jørgen
>
> Best regards,
>
> Marcin
>
>
> On Fri, Jun 17, 2016 at 11:36 AM, Graham Inggs <graham.inggs at gmail.com
> <mailto:graham.inggs at gmail.com>> wrote:
>
> On 17/06/2016 11:28, Ask Hjorth Larsen wrote:
>
> The tests are ordered roughly by how expensive they are to run.
> --from TEST means run all tests after the given one (you want
> to check
> that your fix changed the currently failing test, but you
> don't want
> to run all the unrelated ones that did not fail anyway),
> --after TEST
> means the same but do not include the given one (you want to
> avoid a
> parallel deadlock or segfault which crashed the test suite). --to
> does not exist yet. --range would correspond to both a --from
> and a
> --to, but I think they should just be separate keywords.
>
>
> Aha, I see them all, in order, in gpaw/test/__init__.py
> I should have grepped before asking.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.fysik.dtu.dk/pipermail/gpaw-users/attachments/20160620/03392a3c/attachment.html>
More information about the gpaw-users
mailing list