[petsc-users] PaStiX does not work

Barry Smith bsmith at mcs.anl.gov
Tue Apr 6 21:40:38 CDT 2010

    If you use MUMPS then you should be able to use PaStiX just by  
changing that one line. It should not suddenly say that it is using  
ILU when you have indicated LU.
Are you sure that PETSc was installed to use PasTIX? PasTIX certainly  
will work with KSPREONLY (just like MUMPS).

    Perhaps you should run the entire application with valgrind (www.valgrind.org 
) http://www.mcs.anl.gov/petsc/petsc-as/documentation/ 
faq.html#valgrind to see if there is some memory corruption that is  
causing the problem.

   So run with MUMPS, change that one line and run with PASTIX and  
send us the output (with like -ksp_view) for the two cases all to petsc-maint at mcs.anl.gov 
  not petsc-users


On Apr 6, 2010, at 9:33 PM, Gong Ding wrote:

> Dear Barry
> The mainly petsc related file is attached.
> The whole project is very large, it is a 3D/parallel code for  
> semiconductor simulation.
> You can download the source code from http://www.cogenda.com/downloads/category/7-genius-open-source-edition.html
> It can be compiled with petsc 3.0. I am now adding 3.1 support.
> I had use MUMPS solver for quite a long time by the same code:
> PCFactorSetMatSolverPackage (pc, MAT_SOLVER_MUMPS);
> It works well for most of the time. However, it may crash (when  
> other solver like KSP type works).
> I'd like to try PASTIX if it is faster and more stable.
> Yours
> Gong Ding
>> On Apr 6, 2010, at 7:04 PM, Gong Ding wrote:
>>> Dear Petsc developer,
>>> I found a problem that PASTIX solver can not be loaded from PETSC,
>>> even for 3.1 version.
>>> When I use following code, the ksp_view told me that PC is ilu. And
>>> the snes solver can not convergence at all.
>>> ierr = KSPSetType (ksp, (char*) KSPPREONLY); assert(!ierr);//it
>>> seems PaStiX don't work with KSPPREONLY
>>> ierr = PCSetType  (pc, (char*) PCLU);
>>> ierr = PCFactorSetMatSolverPackage (pc, MAT_SOLVER_PASTIX); assert(!
>>> ierr);
>>> Is it a bug or I made something wrong?
>>  You are doing something wrong but one cannot tell from just this
>> code fragment.  Send the entire code to petsc-maint at mcs.anl.gov and
>> the options you used to run it and the output it produced.
>>   Barry
>>> Yours
>>> Gong Ding
> <fvm_nonlinear_solver.cc>

More information about the petsc-users mailing list