[petsc-dev] Scotch 6.0.1 unifies _esmumps versions

Satish Balay balay at mcs.anl.gov
Sun Feb 15 13:19:43 CST 2015


ops - forgot to list the commit

https://bitbucket.org/petsc/petsc/commits/4875ef09651bdaec9821537d04529c5e45819f80

Satish

On Sun, 15 Feb 2015, Satish Balay wrote:

> Hm - should this go into maint?
> 
> I get the following diff with testexamples_PTSCOTCH
> 
> [I could update the output file with the changed partitioning - assuming its correct. Matt?]
> 
> I have my current changes in branch "balay/update-scotch_6.0.3/maint"
> 
> Satish
> 
> >>>>>>>>>>>
> testexamples_PTSCOTCH in: /home/balay/petsc/src/mat/examples/tutorials
> 4,7c4,7
> < [0] 0 1
> < [0] 1 1
> < [0] 2 1
> < [0] 3 1
> ---
> > [0] 0 2
> > [0] 1 2
> > [0] 2 2
> > [0] 3 0
> 11c11
> < [1] 2 0
> ---
> > [1] 2 1
> 13,15c13,15
> < [2] 0 2
> < [2] 1 2
> < [2] 2 2
> ---
> > [2] 0 1
> > [2] 1 1
> > [2] 2 1
> /home/balay/petsc/src/mat/examples/tutorials
> Possible problem with ex15_2, diffs above
> <<<<<<<<<
> 
> balay at asterix /home/balay/petsc/src/mat/examples/tutorials (balay/update-scotch_6.0.3/maint)
> $ mpiexec -n 3 ./ex15 -mat_partitioning_type ptscotch  -mat_view
> Mat Object: 3 MPI processes
>   type: mpiaij
> row 0: (0, 1)  (1, 1) 
> row 1: (0, 1)  (1, 1)  (2, 1) 
> row 2: (1, 1)  (2, 1)  (3, 1) 
> row 3: (2, 1)  (3, 1)  (4, 1) 
> row 4: (3, 1)  (4, 1)  (5, 1) 
> row 5: (4, 1)  (5, 1)  (6, 1) 
> row 6: (5, 1)  (6, 1)  (7, 1) 
> row 7: (6, 1)  (7, 1)  (8, 1) 
> row 8: (7, 1)  (8, 1)  (9, 1) 
> row 9: (8, 1)  (9, 1) 
> Mat Object: 3 MPI processes
>   type: mpiadj
> row 0: 1 
> row 1: 0  2 
> row 2: 1  3 
> row 3: 2  4 
> row 4: 3  5 
> row 5: 4  6 
> row 6: 5  7 
> row 7: 6  8 
> row 8: 7  9 
> row 9: 8 
> IS Object: 3 MPI processes
>   type: general
> [0] Number of indices in set 4
> [0] 0 2
> [0] 1 2
> [0] 2 2
> [0] 3 0
> [1] Number of indices in set 3
> [1] 0 0
> [1] 1 0
> [1] 2 1
> [2] Number of indices in set 3
> [2] 0 1
> [2] 1 1
> [2] 2 1
> 
> 
> On Sun, 15 Feb 2015, Pierre Jolivet wrote:
> 
> > Hello,
> > Is there a reason why SCOTCH 6.0.0 is still downloaded by PETSc, while 6.0.3 is out and fixes an issue with MUMPS, cf. http://lists.mcs.anl.gov/pipermail/petsc-users/2014-September/022795.html <http://lists.mcs.anl.gov/pipermail/petsc-users/2014-September/022795.html> ?
> > This simple patch seems to work out of the box.
> > 
> > -- a/config/BuildSystem/config/packages/PTScotch.py
> > +++ b/config/BuildSystem/config/packages/PTScotch.py
> > @@ -3,8 +3,7 @@ import config.package
> >  class Configure(config.package.Package):
> >    def __init__(self, framework):
> >      config.package.Package.__init__(self, framework)
> > -    self.download     = ['https://gforge.inria.fr/frs/download.php/31832/scotch_6.0.0_esmumps.tar.gz',
> > -                         'http://ftp.mcs.anl.gov/pub/petsc/externalpackages/scotch_6.0.0_esmumps.tar.gz']
> > +    self.download     = ['http://gforge.inria.fr/frs/download.php/file/34099/scotch_6.0.3.tar.gz']
> > 
> > $ cd src/ksp/ksp/examples/tutorials
> > $ mpirun -np 8 ./ex52 -use_mumps_lu -mat_mumps_icntl_7 5 -m 400 -n 400
> > $ echo $?
> > 0
> > 
> > (this fails with SCOTCH 6.0.0, cf. http://mumps.enseeiht.fr/index.php?page=faq#19 <http://mumps.enseeiht.fr/index.php?page=faq#19>)
> > 
> > Thank you,
> > Pierre
> > 
> > > On Sep 25, 2014, at 6:38 PM, Jed Brown <jed at jedbrown.org> wrote:
> > > 
> > > From na-digest:
> > > 
> > >> From: François Pellegrini francois.pellegrini at labri.fr
> > >> Date: September 20, 2014
> > >> Subject: Revision 6.0.1 of Scotch is out
> > >> 
> > >> We announce the release, as libre/free software, of revision 6.0.1 of
> > >> the SCOTCH and PT-SCOTCH software package and library for sequential
> > >> and parallel graph partitioning, sequential and parallel graph
> > >> clustering, sequential mesh/hypergraph partitioning and static
> > >> mapping, and sequential and parallel sparse matrix block ordering.
> > >> 
> > >> While revision 6.0.1 is technically a bugfix release, with no apparent
> > >> new features, much has changed "under the hood".
> > >> - Sequential algorithms for repartitioning and (re)partitioning with
> > >>  fixed vertices have been improved.
> > >> - Multithreading issues have been solved.
> > >> - From now on, there is no separate "*_esmumps" version for
> > >>  interfacing with the MUMPS solver. The additional libraries can be
> > >>  generated by running "make esmumps" for the sequential libraries and
> > >>  "make ptesmumps" for the parallel libraries.
> > >> 
> > >> Parallel static mapping and parallel direct k-way graph repartitioning
> > >> with fixed vertices will be available in release 6.1.
> > >> 
> > >> Scotch can be freely downloaded, under the terms of the CeCILL-C free
> > >> software license. To ease the development, diffusion, and circulation
> > >> of information regarding the SCOTCH project, most of its resources are
> > >> hosted on the InriaGforge platform provided by INRIA. Please refer to
> > >> the SCOTCH web pages at:
> > >> 
> > >> https://gforge.inria.fr/projects/scotch/
> > >> http://www.labri.fr/~pelegrin/scotch/
> > 
> > 
> 


More information about the petsc-dev mailing list