On Fri, 13 Dec 2013 14:51:19 +0000 "Garth N. Wells" <[email protected]> wrote:
> On 2013-12-13 14:48, Jan Blechta wrote: > > On Fri, 13 Dec 2013 14:40:06 +0000 > > "Garth N. Wells" <[email protected]> wrote: > > > >> On 2013-12-13 14:08, Jan Blechta wrote: > >> > Hi, > >> > > >> > DOLFIN build system looks for scotch_esmumps library. But the > >> > functionality (passing some SCOTCH data to MUMPS elimination > >> > tree computation algorithm) seems to be broken in > >> > scotch_esmumps_6.0.0 according to > >> > http://graal.ens-lyon.fr/MUMPS/index.php?page=faq#19. Does DOLFIN > >> > employ this functionality in fact? Is it then possible to link > >> > DOLFIN with scotch_6.0.0 and esmumps_5.1.12b? > >> > > >> > >> Just use SCOTCH 5.1.12. > >> > >> I loathe to do any special for a package like MUMPS that doesn't > >> bother to keep up with other library releases (MUMPS still does not > >> support ParMETIS 4, despite it being available for 2 1/2 years). > > > > You don't need to do anything - DOLFIN compiles with SCOTCH 6.0.0. > > Moreover MUMPS website claims that this is rather a bug in SCOTCH > > than MUMPS, isn't it? > > > > So they say, but where is the bug report and where is the MUMPS > development repository? Yes, I did not find the bug in SCOTCH bug tracker. > > > Nevertheless you did not answer my question: is this scotch_esmumps > > functionality employed when using DOLFIN? > > > > It is likely required when building DOLFIN with MUMPS. When you're not explicitly aware of this functionality (a communication of the partitioner and the solver) then this is probably not in use. I believe that the solver has no access to partitioning objects to be able to do calls to SCOTCH behind the scene. Am I right? Jan > > Garth > > > Jan > > > >> > >> Garth > >> > >> > Jan > >> > _______________________________________________ > >> > 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
