Hi,

I did something similar for openblas, which seems to work.

See
https://github.com/JaroslavHron/hashstack/blob/master/pkgs/host-openblas.yaml
and
https://github.com/JaroslavHron/hashstack/blob/master/salomon.yaml
(however this compiles its own python)

Jaroslav


On 11/12/2015 10:00 AM, Johannes Ring wrote:
> Hi,
> 
> Using Python from the system is currently only working when Python is
> installed in /usr. I am trying to solve that problem here:
> 
> https://github.com/hashdist/hashstack/pull/883
> 
> Let me know if that fixes the problem for you. Note that you must set
> the HOST_PYTHON parameter in your profile.
> 
> Using OpenBLAS from the system is currently not supported, but it would
> be nice if we could make it work. I did some work to build the FEniCS
> stack with MKL in this branch:
> 
> https://github.com/hashdist/hashstack/tree/johannr/mkl-abel
> 
> Maybe we need to do something similar for using OpenBLAS from the system.
> 
> Johannes
> 
> On Tue, Nov 10, 2015 at 12:24 PM Sven Buijssen
> <[email protected]
> <mailto:[email protected]>> wrote:
> 
>     Hi,
> 
>     I am administrating a few dozen heterogeneous compute nodes running
>     CentOS 6.7
>     64-bit and have been struggling to bend the new build system of
>     FEniCS to our
>     systems' needs since 1.5.0.
>     I had been able to tweak the FEniCS 1.4.0 build system to
>     * not compile ATLAS BLAS,
>     * not compile Python,
>     * not compile Boost,
>     * not compile OpenMPI,
>     * use a newer CGAL and
>     * have all FEniCS prerequisites use the OpenBLAS installation in a
>     non-default path.
> 
>     None of this I have managed with neither 1.5.0 nor 1.6.0. In
>     particular trying
>     to avoid to rebuild Python has proven difficult. I have skimmed over
>     and grepped
>     through the mailing list archive to no avail which is why I am now
>     resorting to
>     contact you.
> 
>     The system heavily relies on Environment Modules and I am trying to
>     avoid that
>     any of the following modules get rebuilt for the sole use of FEniCS
>     1.6.0. In
>     particular, OpenBLAS must not be compiled by FEniCS itself to avoid
>     have a
>     non-relocatable installation (not so much path-relocatable, but in
>     the sense of
>     rsync'able to other servers with different CPUs and cache sizes where an
>     appropriate OpenBLAS installation is readily available already):
> 
>     -------------------- /sfw/Modules/modulefiles ---------------------
>     cmake/2.8.12                      cmake/3.2.2
> 
>     boost/1.41-centos boost/1.48      boost/1.56
> 
>     gcc/4.4.7-centos                  gcc/4.9.2
>     gcc/4.7.3                         gcc/5.1.0
>     gcc/4.8.5                         gcc/5.2.0
> 
>     openmpi/gcc4.4.x/1.8.2            openmpi/intel13.1.x/1.8.5
>     openmpi/gcc4.4.x/1.8.5            openmpi/intel14.0.x/1.8.5
>     openmpi/gcc4.4.x-ib-gnu/1.8.5     openmpi/intel15.0.x/1.8.5
>     openmpi/gcc4.9.x/1.8.5            openmpi/intel15.0.x-ib-gpu/1.10.1
>     openmpi/gcc5.1.x/1.8.5
> 
>     openblas/0.2.13(deprec)           openblas/0.2.15
> 
>     python/2.7.8
> 
> 
>     I have already customized a *.yaml script based on what I could find
>     on the
>     mailing list. I am wondering whether you could give me some pointers
>     as to where
>     I set up the script attached wrongly. (I run it via 'bash; source
>     install-script.wip') For the last two days, I am stuck at the FEniCS
>     1.6.0
>     installation trying to build python - although I thought I had
>     forbidden that in
>     install-script.wip - and failing to locate /usr/bin/python-config
>     (which does
>     indeed not exist there, but in /sfw/python/2.7.8/bin/python-config)
> 
>     Thank you in advance for taking the time to answer me and possibly
>     provide me
>     with hints.
> 
>     With kind regards,
>     Sven Buijssen
> 
> 
> 
> 
>     --
>     Dipl.Math. Sven H.M. Buijssen
>     wiss.Ang./MRB/LiDO-Team
>     Technische Universität Dortmund
>     Institut für Angewandte Mathematik (LS III)
>     Fakultät für Mathematik
>     Vogelpothsweg 87
>     D-44227 Dortmund
> 
>     Tel.: +49 231-755 59 34
>     Fax: +49 231-755 59 33
>     [email protected] <mailto:[email protected]>
>     www.tu-dortmund.de <http://www.tu-dortmund.de> 
>      www.mathematik.tu-dortmund.de/lsiii
>     <http://www.mathematik.tu-dortmund.de/lsiii>
>     www.mathematik.tu-dortmund.de/~buijssen
>     <http://www.mathematik.tu-dortmund.de/~buijssen>
> 
>     Important note: The information included in this e-mail is
>     confidential. It is
>     solely intended for the recipient. If you are not the intended
>     recipient of this
>     e-mail please contact the sender and delete this message. Thank you.
>     Without prejudice of e-mail correspondence, our statements are only
>     legally
>     binding when they are made in the conventional written form (with
>     personal
>     signature) or when such documents are sent by fax.
>     _______________________________________________
>     fenics-support mailing list
>     [email protected]
>     <mailto:[email protected]>
>     http://fenicsproject.org/mailman/listinfo/fenics-support
> 
> 
> 
> _______________________________________________
> fenics-support mailing list
> [email protected]
> http://fenicsproject.org/mailman/listinfo/fenics-support
> 

_______________________________________________
fenics-support mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics-support

Reply via email to