search

ARudik / feelpp

Automatically exported from code.google.com/p/feelpp
0 stars 0 forks source link

Wrong number of dof in parallel ? #45

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago 
Hi

I wrote an e-mail on the mailing list but nobody received it, so I write here.

When running my code on two processors I have a problem when using the boost 
parameter function newMatrix from backendpetsc.
More precisely, to create the matrix,  trial->nLocalDofWithoutGhost() and 
test->nLocalDofWithoutGhost() are used and it seems that my problem comes from 
here. 

First of all I have a question :
If on proc 0 i have 476 dofs, and 480 dofs on proc 1, should i have :
test->nLocalDofWithoutGhost() = 476 ( proc 0 )
trial->nLocalDofWithoutGhost() = 476 ( proc 0 )
and
test->nLocalDofWithoutGhost() : 480 ( proc 1 )
trial->nLocalDofWithoutGhost() : 480 ( proc 1 )
?

Although on proc 0 these numbers are correct, but they are wrong on proc 1 
because I get 465 instead of 480, so on proc 1 my matrix have a wrong size and 
it's annoying to perform matrix-vector multiplications...

Is anyone else have to deal with the same problem ?

Bests,

Original issue reported on code.google.com by stephane...@gmail.com on 7 Aug 2012 at 8:32

GoogleCodeExporter commented 9 years ago 
In feel++, you have 2 sizes : nLocalDofWithoutGhost and nLocalDofWithGhost. For 
Petsc Matrix and Vector in //, you must give nLocalDofWithoutGhost() for the 
localSize
and also a map which take into account the ghosts dofs. Also, in //, it's very 
important
to use newVector(Xh) and newMatrix(Xh,Xh), others constructors don't work in 
parallel yet.
If you use the good constructor, you can send me a simple exemple.

Vincent.

Original comment by vincent....@gmail.com on 7 Aug 2012 at 10:04

GoogleCodeExporter commented 9 years ago 
I ran feel_doc_laplacian on two procs.
The total number of dofs is 65
At the boost parameter functions newMatrix from backend.hpp, i have :
trial->nLocalDofWithoutGhosts() : 36 and test->nLocalDofWithoutGhosts : 36 ( 
proc 0 )
trial->nLocalDofWithoutGhosts() : 29 and test->nLocalDofWithoutGhosts : 29 ( 
proc 1 )
the i have 65 dofs in total, everything is ok unlike that i said yesterday ( 
sorry )

But concerning the element_type u (solution) i have :
u.localSize() : 36 on each procs, but I expected to have 29 on proc 1 ( isn't 
it ? )

and so, if you add following line, after the problem is solved :
D->energy( u , u );
then there is a problem when MatMult is called inside the energy function, and 
i get the following error message :
[1]PETSC ERROR: --------------------- Error Message 
------------------------------------
[1]PETSC ERROR: Nonconforming object sizes!
[1]PETSC ERROR: Mat mat,Vec y: local dim 29 36!
[1]PETSC ERROR: 
------------------------------------------------------------------------
[1]PETSC ERROR: Petsc Release Version 3.2.0, Patch 5, Sat Oct 29 13:45:54 CDT 
2011 

By the way i noticed that it's not a good idea to compile feel_doc_laplacian 
with clang, the compilation is ok but then it crashes whereas everything is ok 
whith gcc-4.5 or gcc-4.6

Original comment by stephane...@gmail.com on 8 Aug 2012 at 7:48

GoogleCodeExporter commented 9 years ago 
energy( ) funtion in matrixpetsc.hpp is not //.
you try to use MatMult with a parallel matrix and a sequential vector.
To fix that you should implement energy function in MatrixPetscMPI
and used VectorPetscMPI !! (not VectorPetsc which work only in sequential and 
idem for MatrixPetsc and MatrixPetscMPI).

To build properly a VectorPetscMPI, you should use this constructor 
VectorPetscMPI( DataMap const& dm )
with dm = this->mapRow() ( keyword "this" is a MatrixPetscMPI).

Original comment by vincent....@gmail.com on 8 Aug 2012 at 8:32

GoogleCodeExporter commented 9 years ago 
Thanks Vincent for your advices
I added energy function implementation in parallel but unfortunately I don't 
obtain same result than in sequential so I still have a bug to find ...

Original comment by stephane...@gmail.com on 9 Aug 2012 at 7:00

GoogleCodeExporter commented 9 years ago 
should this bug be closed now ?

Original comment by christop...@feelpp.org on 10 Aug 2012 at 11:10

GoogleCodeExporter commented 9 years ago 
yes

Original comment by stephane...@gmail.com on 10 Aug 2012 at 5:01