[petsc-dev] Problematic Merge of FieldSplit

[Patrick Lacasse]

> On Jul 8, 2012, at 3:36 PM, Dmitry Karpeev wrote:
>>
>> > As currently designed, there are always two KSPs for A00 - one is
>> created as part of MatSchurComplement and the other one sits in the FS
>> split.
>>
>>    Really? Isn't that often non-efficient? For any expensive to build and
>> memory intensive preconditioner (like ILU(k), LU, AMR) it is crazy to
>> create them both if they are the same.
>>
> Actually, until recently these  two KSPs had different options prefixes:
> the outer had -fieldsplit_0 and the inner -fieldsplit_1,
> so if they were set up from the command line, they could (and did for me)
> end up being different, so this reuse of the KSP would be a new thing.
>  Reusing the factors from ILU(k) or LU would be a good thing, since the
> same solve is done as many as 3 times sometimes.  The same would be true
> for an ASM-type preconditioner with (I)LU on the blocks -- even more so,
> since submatrices are getting pulled out.
>
>>
>>
Could a  PCFieldSplitSetSubKSP be usefull ? So that I could take the ksp
from outer A00 and give it to schur A00.
I'm using a LU factorisation for A00 and I must reuse the factorisation.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20120709/6aa4d12c/attachment.html>