[petsc-dev] MatSetValuesStencil analogue in petsc4py

[Chris Eldred]

Hey Nathan,

Thank you- that example is perfect, looks like exactly what we want to
do. Not sure how I missed it!

-Chris

On Thu, Nov 19, 2015 at 1:46 PM, Nathan Collier
<nathaniel.collier at gmail.com> wrote:
> Chris,
>
> I believe Lisandro does have an example that can help you here. Take a look
> at his demo here (petsc4py/demo/wrap-cython/):
>
> https://bitbucket.org/petsc/petsc4py/src/2c8d5dcc0a7fb9519e68d3caba5a0501f4bb6143/demo/wrap-cython/?at=master
>
> He implements a FormFunction and FormInitialGuess in pure C, wraps them with
> cython and then calls them from python. Is this what you had in mind?
>
> Nate
>
>
>
> On Thu, Nov 19, 2015 at 3:41 AM Chris Eldred <chris.eldred at gmail.com> wrote:
>>
>> Hey Jed,
>>
>> I will give that a look today and see what I can do.
>>
>> On a more general note, what I would really like to be able to do is
>> use petsc4py as a driver script that just calls PETSc functions, and
>> do the computationally expensive parts such as stiffness matrix
>> assembly and load vector assembly in C.
>>
>> To this end, is it possible to create a PETSc matrix or vector in
>> Python, and then pass them to a C function that modifies them? Then
>> MatSetValues would be called in C, but the AssemblyBegin/AssemblyEnd
>> pairs (and Matrix creation, etc) could be called in Python.
>>
>> -Chris
>>
>> On Wed, Nov 18, 2015 at 8:27 PM, Jed Brown <jed at jedbrown.org> wrote:
>> > Nathan Collier <nathaniel.collier at gmail.com> writes:
>> >> ack MatSetValuesStencil
>> >>
>> >> returns
>> >>
>> >> src/PETSc/petscmat.pxi
>> >> 222:    int
>> >>
>> >> MatSetValuesStencil(PetscMat,PetscInt,PetscMatStencil[],PetscInt,PetscMatStencil[],PetscScalar[],PetscInsertMode)
>> >> 1027:        CHKERR( MatSetValuesStencil(A,
>> >>
>> >> but 'ack MatSetValueStencil' returns nothing. Hope it helps,
>> >
>> > Mat.pyx defines Mat.setValueStencil which calls petscmat.pxi's
>> > matsetvaluestencil which calls PETSc's MatSetValuesStencil with an array
>> > of length 1.  I can't think of any reason not to generalize this to set
>> > many values at once.  Chris, would you like to take a pass at adding
>> > this?
>>
>>
>>
>> --
>> Chris Eldred
>> Postdoctoral Fellow, LAGA, University of Paris 13
>> PhD, Atmospheric Science, Colorado State University, 2015
>> DOE Computational Science Graduate Fellow (Alumni)
>> B.S. Applied Computational Physics, Carnegie Mellon University, 2009
>> chris.eldred at gmail.com



-- 
Chris Eldred
Postdoctoral Fellow, LAGA, University of Paris 13
PhD, Atmospheric Science, Colorado State University, 2015
DOE Computational Science Graduate Fellow (Alumni)
B.S. Applied Computational Physics, Carnegie Mellon University, 2009
chris.eldred at gmail.com