[petsc-dev] To add a PETSc dependent package to PETSc as an external package.

[Xuefei (Rebecca) Yuan]

Dear Sherry,

I have uploaded the tarball at

http://portal.nersc.gov/project/nerscpd/pdslin_0.0.tar.gz

and I asked Ichi about if this is the current version we have on PDSLin.

Best regards,

Rebecca


On Aug 17, 2011, at 2:14 PM, Xiaoye S. Li wrote:

> Ichi,
> Can you please sned a tarball of the latest code to petsc-maint?
> 
> Barry,
> Please install the hook to PETSc's development branch, but not for distribution.  We are still waiting for the LBL's mBSD license before release. 
> 
> Thanks,
> Sherry
> 
> 
> On Wed, Aug 17, 2011 at 1:46 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
>   Rebecca,
> 
>    It would probably be best if you email to petsc-maint the PDSLin tarball so we can look at the PDSlin structure and come up with the best way to proceed. The way to handle the packaging will depend on how PDSLin interacts with PETSc.
> 
>   Thanks
> 
>   Barry
> 
> On Aug 17, 2011, at 3:19 PM, Xuefei (Rebecca) Yuan wrote:
> 
> > Dear all,
> >
> > We are trying to add a PETSc dependent package called PDSLin to PETSc as an external package. PDSLin is a hybrid linear solver and we would like to have it been imported to PETSc as SuperLU_DIST used to be, and it provides an alternative way to solving a linear system in parallel.
> >
> > Even though the package has not been released yet (should be released soon) and it is under testing for some problems, we are still interesting in putting it as an external package to PETSc for now.
> >
> > I did some search of the old threads to the petsc-dev team and there is a request from Lious for the PaStiX a few years ago.
> >
> > My understanding to add a package is the following steps:
> >
> > 1) create a python code called "PDSLin.py" under /petsc-dev/config/PETSc/packages
> >
> > 2) write an interface file between PETSc and PDSLin.
> >
> > As PDSLin is very similar to SuperLU_DIST, and it also depending on PETSc (similar to Prometheus), the PDSLin.py is created based on SuperLU_DIST.py and Prometheus.py.
> >
> > Before putting the interface file in some places under petsc-dev directory, would you please help me understand how those option calls go to this interface file?
> >
> > a) am I missing another file to create/modify to have those options connected to PDSLin in a very similar way to SuperLU_DIST?
> >
> > b) do I have some flexibility to add more options called from PETSc to this package?
> >
> > Moreover, I am not sure how to put the interface file in the dev package.
> >
> > c) where shall it go? The SuperLU_DIST is located at
> >
> > ./src/mat/impls/aij/mpi/superlu_dist/superlu_dist.c
> >
> > shall I put it as
> >
> > ./src/mat/impls/aij/mpi/pdslin/pdslin.c  ?
> >
> > d) what is the rule of writing this interface file?
> >
> > In superlu_dist.c, the extern functions are
> >
> > extern PetscErrorCode MatFactorInfo_SuperLU_DIST(Mat,PetscViewer);
> > extern PetscErrorCode MatLUFactorNumeric_SuperLU_DIST(Mat,Mat,const MatFactorInfo *);
> > extern PetscErrorCode MatDestroy_SuperLU_DIST(Mat);
> > extern PetscErrorCode MatView_SuperLU_DIST(Mat,PetscViewer);
> > extern PetscErrorCode MatSolve_SuperLU_DIST(Mat,Vec,Vec);
> > extern PetscErrorCode MatLUFactorSymbolic_SuperLU_DIST(Mat,Mat,IS,IS,const MatFactorInfo *);
> > extern PetscErrorCode MatDestroy_MPIAIJ(Mat);
> >
> > What kind of functions should I have in the interface file for PDSLin?
> >
> > I am not sure if I have stated the problem clearly, in short is that we would like to add some package to PETSc but not sure the right way and steps to approach the goal.
> >
> > Thanks very much!
> >
> > Best regards,
> >
> > Rebecca
> >
> > Xuefei (Rebecca) Yuan
> > Postdoctoral Fellow
> > Lawrence Berkeley National Laboratory
> > Tel: 1-510-486-7031
> >
> >
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20110817/939a8d8c/attachment.html>