[petsc-users] some questions about PETSc

[Aron Ahmadia]

Thanks for the extra notes Lisandro, I've migrated some of the results of
this discussion to the petclaw development wiki here:
https://bitbucket.org/knepley/petclaw/wiki/How_do_I

A

On Sat, Sep 18, 2010 at 8:55 PM, Lisandro Dalcin <dalcinl at gmail.com> wrote:

> On 18 September 2010 21:01, Aron Ahmadia <aron.ahmadia at kaust.edu.sa>
> wrote:
> > Dear Amal,
> > Thanks for the questions.  These are great!  I think they show a good
> > fundamental approach, you are thinking about these problems like a PETSc
> > scientist would.  I am going to cc petsc-users on the reply in case
> anybody
> > wants to add or comment:
> > How to set up a DA for multiple equations in petsc4py or PETSc? (for q)
> > The PETSc terminology for the number of equations active at a given node
> is
> > 'degrees of freedom', which as far as I can tell, is a term borrowed from
> > finite element analysis of structures, originally referring (in
> mechanics)
> > to the number of potential displacements or rotations that specify an
> > element in the system, but in computational science as a way to describe
> the
> > total number of equations in the discretized system (i.e., the number of
> > elements in the right-hand side vector).  After you have called DACreate,
> > you call DASetDof
> > (
> http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/docs/manualpages/DA/DASetDof.html
> )
> > to set the number of degrees of freedom per vertex.
>
> In petsc4py, you pass "dof" as a keyword argument to create() method:
>
> da = PETSc.DA().create((Nx, Ny, Nz), dof=4, ...)
>
>
> > How to deal with boundary conditions where you have special treatment for
> > them?  In an example we have seen that is done by simply if statements
> that
> > checks if this point is a boundary condition cell, should we do it this
> way.
> > As far as I know, this is the PETSc way, so you do not need to do
> anything
> > differently :o)  Later on, when we are tuning for performance, we might
> > extract the boundary-handling code from the main for loop, and create
> index
> > sets to represent each of the sections of the array we would like to
> iterate
> > over.  If we profile the code and this is a non-significant section, then
> we
> > just leave it.
>
> And with petsc4py running on parallel, this is a mess to implement. I
> still have to found a way to emulate DA Vec arrays, as NumPy arrays
> does not support lower bounds in dimensions.
>
> > When we start incorporating petsc4py structures in clawpack, output is
> being
> > duplicate for each process, is the way to avoid this is by (if
> statements)
> > to determine that everything need to be done once is the responsibility
> of
> > process 0 for instance, or there are other ways to do it?
> > PetscPrintf
> > (
> http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-dev/docs/manualpages/Sys/PetscPrintf.html
> )
> > will do what you want.
>
> In petsc4py, you can use PETSc.Sys.Print (also look at
> PETSc.Sys.syncPrint and PETSc.Sys.syncFlush). PETSc.Sys.Print (and
> syncPrint) should be used like print function Python 3, but it
> interprets an extra 'comm'  kw argument.
>
>
>
> --
> Lisandro Dalcin
> ---------------
> CIMEC (INTEC/CONICET-UNL)
> Predio CONICET-Santa Fe
> Colectora RN 168 Km 472, Paraje El Pozo
> Tel: +54-342-4511594 (ext 1011)
> Tel/Fax: +54-342-4511169
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20100918/67092959/attachment.htm>