seg fault at test run on CentOS
Hello, I have installed the latest version of kwant on a CentOS system that I do not have root access. So I installed it on my home directory. When I run the test file, i get the following seg fault. python3 -c 'import kwant; kwant.test(verbose=2)' /lib/python3.6/site-packages/kwant-unknown-py3.6-linux-x86_64.egg/kwant/solvers/default.py:18: RuntimeWarning: MUMPS is not available, SciPy built-in solver will be used as a fallback. Performance can be very poor in this case. "Performance can be very poor in this case.", RuntimeWarning) kwant.graph.tests.test_core.test_empty ... ok kwant.graph.tests.test_core.test_num_nodes ... ok .... kwant.tests.test_lattice.test_neighbors ... ok kwant.tests.test_lattice.test_shape ... ok kwant.tests.test_lattice.test_wire ... ok kwant.tests.test_lattice.test_translational_symmetry ... Segmentation fault Where I skipped a lot of lines, they were ok for most, skipped when related to MUMPS. I am not that familiar with python or kwant to know what this error means. Can someone help? Thanks Fatih
Hi Faith,
I would suggest to try to install kwant via conda.
With: conda install -c conda-forge kwant
Best, Bas
On 27 Feb 2017, 11:55 +0100, Fatih Dogan
Hello,
I have installed the latest version of kwant on a CentOS system that I do not have root access. So I installed it on my home directory. When I run the test file, i get the following seg fault.
python3 -c 'import kwant; kwant.test(verbose=2)' /lib/python3.6/site-packages/kwant-unknown-py3.6-linux-x86_64.egg/kwant/solvers/default.py:18: RuntimeWarning: MUMPS is not available, SciPy built-in solver will be used as a fallback. Performance can be very poor in this case. "Performance can be very poor in this case.", RuntimeWarning) kwant.graph.tests.test_core.test_empty ... ok kwant.graph.tests.test_core.test_num_nodes ... ok .... kwant.tests.test_lattice.test_neighbors ... ok kwant.tests.test_lattice.test_shape ... ok kwant.tests.test_lattice.test_wire ... ok kwant.tests.test_lattice.test_translational_symmetry ... Segmentation fault
Where I skipped a lot of lines, they were ok for most, skipped when related to MUMPS. I am not that familiar with python or kwant to know what this error means. Can someone help?
Thanks
Fatih
Hi, To expand on Bas' answer I believe that this is related to the fact that somehow Kwant has not been linked correctly against system libraries. Indeed Bas' solution to use 'conda' may be the quickest way to get kwant working (if you already have a working Anaconda installation), however we would still like to find the root cause of this. Can you give details of how you built/installed Kwant on your system? Regards, Joe
Thank you for the replies. I do not have Anaconda installed. It appears to be a major software (400MB size). System has python 2.6 installed. So I began with installing python 3.6. I installed pip and all dependencies (Scipy etc) on my home folder. For Kwant, the only change I made to the tar.gz is to change build.conf to include lapack library i compiled using gcc. [lapack] libraries = lapack refblas library_dirs = . extra_link_args = -Wl -lgfortran # I hope this helps. Regards, Fatih
Hi,
I do not have Anaconda installed. It appears to be a major software (400MB size).
Indeed Anaconda is a large package. There is, however, a stripped down version called Miniconda [1], which installs just the "conda" tool and a version of Python (rather than the whole scientific Python stack). The Miniconda installer is a more reasonable size (40MB). The advantage of using conda over, say, pip is that there are conda packages for most non-python dependencies required for scientific computing, which means that there are fewer problems when it comes to linking, as conda knows about the whole software stack.
[lapack] libraries = lapack refblas library_dirs = . extra_link_args = -Wl -lgfortran
I am almost certain that this is a linking issue, e.g. Kwant is linked against some version of BLAS and LAPACK is linked against another, or something like that. Without being able to exactly reproduce your setup I probably won't be able to say any more than that. I would strongly suggest using conda; this will probably get you set up the fastest. Regards, Joe [1]: https://conda.io/miniconda.html
Hello,
Thank you very much. miniconda worked. I had to manually remove everything
from my account and install from scratch, but i got it to work (for testing
at least). The only skip I got was:
kwant.solvers.tests.test_sparse.test_umfpack_del ... SKIP
which i wont bother to check for now.
Please expect more questions as i dig into kwant more.
Thanks again.
Fatih
On Mon, Feb 27, 2017 at 4:26 PM, Joseph Weston
Hi,
I do not have Anaconda installed. It appears to be a major software (400MB size).
Indeed Anaconda is a large package. There is, however, a stripped down version called Miniconda [1], which installs just the "conda" tool and a version of Python (rather than the whole scientific Python stack). The Miniconda installer is a more reasonable size (40MB). The advantage of using conda over, say, pip is that there are conda packages for most non-python dependencies required for scientific computing, which means that there are fewer problems when it comes to linking, as conda knows about the whole software stack.
[lapack] libraries = lapack refblas library_dirs = . extra_link_args = -Wl -lgfortran
I am almost certain that this is a linking issue, e.g. Kwant is linked against some version of BLAS and LAPACK is linked against another, or something like that. Without being able to exactly reproduce your setup I probably won't be able to say any more than that.
I would strongly suggest using conda; this will probably get you set up the fastest.
Regards,
Joe
Hi,
Thank you very much. miniconda worked. I had to manually remove everything from my account and install from scratch, but i got it to work (for testing at least).
Really? Usually conda keeps everything pretty separate; you can also activate different environments similar to the python-only "virtualenv".
The only skip I got was:
kwant.solvers.tests.test_sparse.test_umfpack_del ... SKIP
which i wont bother to check for now.
Yes this is known behaviour. I thought we had removed the UMFPACK code, so I'm not sure why this is still there.
Please expect more questions as i dig into kwant more.
Happy Kwanting! Joe
participants (3)
-
Bas Nijholt
-
Fatih Dogan
-
Joseph Weston