Yes, looks like we need to use KSPSetReusePreconditioner with new PETSc 
(http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/KSP/KSPSetReusePreconditioner.html).
 

Mikael

> On 21 Mar 2015, at 21:50, Garth N. Wells <[email protected]> wrote:
> 
> On Sat, Mar 21, 2015 at 7:32 PM, Mikael Mortensen
> <[email protected]> wrote:
>> I see, but the call to solve also calls set_operator, so the code on the Q&A 
>> should work as far as I can tell? Why is "different_nonzero_pattern" faster 
>> than "same”? It used to (and should) be the other way around.
>> 
> 
> The difference (2.43 vs 2.32) is too small to be significant, at least
> without averaging over a number of runs.
> 
> The options "different_nonzero_pattern" and "same” were removed from PETS 3.5.
> 
> Garth
> 
>> Mikael
>> 
>>> On 21 Mar 2015, at 20:09, Garth N. Wells <[email protected]> wrote:
>>> 
>>> Call ‘set_operator' when you want the preconditioner to be updated.
>>> 
>>> Garth
>>> 
>>> 
>>>> On 21 Mar 2015, at 18:54, Mikael Mortensen <[email protected]> 
>>>> wrote:
>>>> 
>>>> Hi,
>>>> 
>>>> Seems like there is a problem with preconditioner reuse in dolfin 1.5: 
>>>> http://fenicsproject.org/qa/6817/reuse-of-krylov-preconditioner?show=6838#c6838
>>>> 
>>>> Is this a bug? I can only see this issue that is marked as resolved: 
>>>> https://bitbucket.org/fenics-project/dolfin/issue/269/update-logic-on-petsc-preconditioner-re
>>>>  . The function set_petsc_operators that used to set the preconditioner 
>>>> structure is commented out in PETScKrylovSolver.cpp...
>>>> 
>>>> Regards
>>>> 
>>>> Mikael
>>>> 
>>>> 
>>>> _______________________________________________
>>>> fenics mailing list
>>>> [email protected]
>>>> http://fenicsproject.org/mailman/listinfo/fenics
>>> 
>> 

_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics

Reply via email to