[Debichem-devel] current status of mpi-default-dev for mpich/mpich2 arches (and scalapack in particular)

Sébastien Villemot sebastien at debian.org
Tue Mar 4 13:44:23 UTC 2014


Le mardi 04 mars 2014 à 14:40 +0100, Michael Banck a écrit :
> On Tue, Mar 04, 2014 at 02:37:28PM +0100, Sébastien Villemot wrote:
> > Le mardi 04 mars 2014 à 13:18 +0100, Michael Banck a écrit :
> > > cp2k FTBFS on s390x[1], because:
> > > 
> > > 1. The content of /usr/share/mpi-default-dev/debian_defaults on
> > > mpich-arches is currently "mpich" in unstable, TTBOMK.
> > > 
> > > 2. The scalapack (and blacs, I believe) libraries contain a
> > > "libscalapack-mpich2.so" symlink.
> > > 
> > > 3. In DebiChem, we have several packages which link to scalapack, and
> > > sofar we used something like this snippet to figure out which library to
> > > link in:
> > > include /usr/share/mpi-default-dev/debian_defaults
> > > export LIB_SCALAPACK=-lscalapack-$(ARCH_DEFAULT_MPI_IMPL)
> > 
> > Note that I have requested a binNMU on scalapack (#740620) which should
> > solve this specific issue. petsc is currently affected by the same
> > problem as cp2k.
> > 
> > I don't know if we need to think of a better solution to this problem.
> > As long as mpich does not get renamed again in the foreseeable future,
> > this may not be necessary.
> 
> Ah great - I didn't realize a binNMU would fix this.

Indeed I could verify on zelenka that after a rebuild, the scalapack
libraries will have the "-mpich" suffix (instead of "-mpich2").

Don't hesitate to mention in the binNMU bug that this issue is also
affecting cp2k.

-- 
 .''`.    Sébastien Villemot
: :' :    Debian Developer
`. `'     http://www.dynare.org/sebastien
  `-      GPG Key: 4096R/381A7594

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part
URL: <http://lists.alioth.debian.org/pipermail/debichem-devel/attachments/20140304/4ae5b03b/attachment.sig>


More information about the Debichem-devel mailing list