[petsc-dev] MatLUFactor is an orphan

[Jed Brown]

On Sat, Oct 20, 2012 at 6:17 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:

>
>   I believe MatLUFactor() is pretty bogus anyways. In place LU
> factorization only makes sense for dense matrices (LAPACK and LINPACK).
>
>    This is currently an interface to Elemental for MatLUFactor().
>
>   Should we change it to MatDenseLUFactor()?  No solver package business
> needed separately since Elemental is its own matrix type anyway.
>

Meh, I don't have a problem leaving it, but maybe the out-of-place version
like in MatLUFactor_SeqAIJ() should be pushed up to MatLUFactor() so it can
do something reasonable when the function pointer isn't set.

I was just annoyed that a formerly serial example didn't run in parallel
because it called MatLUFactor.


>
>    Barry
>
>
> On Oct 20, 2012, at 4:57 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
>
> > The dispatch mechanism is totally different from
> MatLUFactorSymbolic/Numeric, with no ability to choose the factorization
> package. Should it be changed to take a MatSolverPackage and then to
> dispatch using a MatGetFactor?
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20121020/2316c447/attachment.html>