On Wednesday 18 February 2009 11:03:15 Johan Hake wrote: > On Wednesday 18 February 2009 11:00:36 Garth N. Wells wrote: > > Johan Hake wrote: > > > On Wednesday 18 February 2009 10:23:52 Garth N. Wells wrote: > > >> Johan Hake wrote: > > >>> On Wednesday 18 February 2009 09:38:05 DOLFIN wrote: > > >>>> One or more new changesets pushed to the primary dolfin repository. > > >>>> A short summary of the last three changesets is included below. > > >>>> > > >>>> changeset: 5728:c01d60f80a759d8126dd67471db65d9b733ff1ee > > >>>> tag: tip > > >>>> user: "Johan Hake <[email protected]>" > > >>>> date: Wed Feb 18 09:37:59 2009 +0100 > > >>>> files: dolfin/swig/dolfin_la_post.i > > >>>> test/unit/la/python/test.py description: > > >>>> Fix bug in numpy wrapping of CRS data. > > >>>> - Numpy type needed to be NPY_ULONG instead of NPY_UINT(!?) > > >>> > > >>> I noticed a qurious thing while fixing this bug. The sparsity pattern > > >>> are not the same at a 64 bit computer and on my laptop, a 32 bit one. > > >>> > > >>> Any clue why? > > >> > > >> Strange. Can you add some simple code to the sandbox to which > > >> reproduces this? > > > > > > Done. > > > > > > I have checked some more and it turns out that it only applies for the > > > MTL4 backend. It removes the nonzero values from the pattern on my > > > laptop. > > > > Are you looking at the sparsity pattern or the assembled matrix? The > > MTL4 wrappers cheat a bit and don't use information from the sparsity > > pattern other than the matrix dimensions. > > On the assembled matrix. > > Johan > > > Garth > > > > > I also have different versions of MTL4 on the two machines: > > > > > > 64 bit: svn revision 6624 > > > 32 bit: svn revision 6597
It also turned out to that when I updated the MTL4 on my laptop there were no cheating in the MTL4 library anymore. Johan > > > > > > Could be this. > > > > > > Johan > > _______________________________________________ > DOLFIN-dev mailing list > [email protected] > http://www.fenics.org/mailman/listinfo/dolfin-dev _______________________________________________ DOLFIN-dev mailing list [email protected] http://www.fenics.org/mailman/listinfo/dolfin-dev
