[petsc-users] Issue to set values to a matrix in parallel in python

[Matthew Knepley]

On Tue, Aug 9, 2022 at 11:12 AM Stefano Zampini <stefano.zampini at gmail.com>
wrote:

> PETSc distributes matrices and vectors in parallel. Take a look at
> https://petsc.org/release/docs/manualpages/Vec/VecGetOwnershipRange.html


Longer explanation:

Vectors are distributed by giving a stretch of contiguous rows to each
process. Dense matrices also give a stretch of contiguous rows to each
process.
This means that each matrix column would correspond to a vector. The
easiest way I see to do this is to use

  https://petsc.org/main/docs/manualpages/Mat/MatDenseGetColumnVec/

and then VecCopy() each vector into the correct column. If you want to
share memory, allocate the dense matrix first, and make you vectors using
this
routine.

  Thanks,

      Matt


>
> > On Aug 9, 2022, at 11:21 AM, Thomas Saigre <saigre at math.unistra.fr>
> wrote:
> >
> > Hi,
> >
> > I've been trying for a few weeks to construct a matrix from a list of
> vectors, unsuccessfully.
> > Here is my problem : I have a list l a petsc4py.Vec, each vector has a
> size n, and I have d vectors. I want to "cast" these vectors to a
> petsc4py.Mat Z of shape (n,d), where Z[:, i] = l[i] (using NumPy notation)
> >
> > Here is the code I'm using :
> >
> > import sys
> > from petsc4py import PETSc
> > n = 5
> > d = 10
> >
> > l = []   # creation of the list of vectors
> > for i in range(d):
> >     v = PETSc.Vec().create()
> >     v.setSizes(n)
> >     v.setFromOptions()
> >     v.set(i)
> >     l.append(v)
> >
> > Z = PETSc.Mat().create()
> > Z.setSizes([n, d])
> > Z.setFromOptions()
> > Z.setUp()
> > for i, v in enumerate(l):
> >     Z.setValues(range(n), i, v)
> > Z.assemble()
> > Z.view()    # to display the matrix in the terminal
> >
> > In sequential, the result is correct :
> >
> > Mat Object: 1 MPI processes
> >   type: seqaij
> > row 0: (0, 0.)  (1, 1.)  (2, 2.)  (3, 3.)  (4, 4.)  (5, 5.)  (6, 6.)
> (7, 7.)  (8, 8.)  (9, 9.)
> > row 1: (0, 0.)  (1, 1.)  (2, 2.)  (3, 3.)  (4, 4.)  (5, 5.)  (6, 6.)
> (7, 7.)  (8, 8.)  (9, 9.)
> > row 2: (0, 0.)  (1, 1.)  (2, 2.)  (3, 3.)  (4, 4.)  (5, 5.)  (6, 6.)
> (7, 7.)  (8, 8.)  (9, 9.)
> > row 3: (0, 0.)  (1, 1.)  (2, 2.)  (3, 3.)  (4, 4.)  (5, 5.)  (6, 6.)
> (7, 7.)  (8, 8.)  (9, 9.)
> > row 4: (0, 0.)  (1, 1.)  (2, 2.)  (3, 3.)  (4, 4.)  (5, 5.)  (6, 6.)
> (7, 7.)  (8, 8.)  (9, 9.)
> >
> >  but when I run it using the command mpirun -np 2 python3 file.py, I get
> the following error, about incompatible array sizes (I did not manage to
> understand what ni, nj and nv correspond to...)
> >
> > Traceback (most recent call last):
> >   File "/home/Documents/code/tests/file.py", line 31, in <module>
> >     Z.setValues(list(range(n)), i, v)
> >   File "PETSc/Mat.pyx", line 888, in petsc4py.PETSc.Mat.setValues
> >   File "PETSc/petscmat.pxi", line 828, in petsc4py.PETSc.matsetvalues
> > ValueError: incompatible array sizes: ni=5, nj=1, nv=3
> > Traceback (most recent call last):
> >   File "/home/saigre/Documents/code/tests/t2.py", line 31, in <module>
> >     Z.setValues(list(range(n)), i, v)
> >   File "PETSc/Mat.pyx", line 888, in petsc4py.PETSc.Mat.setValues
> >   File "PETSc/petscmat.pxi", line 828, in petsc4py.PETSc.matsetvalues
> > ValueError: incompatible array sizes: ni=5, nj=1, nv=2
> >
> > Two weeks ago, I made a post on stack overflow (
> https://stackoverflow.com/questions/73124230/convert-a-list-of-vector-to-a-matrix-with-petsc4py).
> I tried using the apt packages, and I also compiled from the sources, but I
> get the same error.
> >
> > I someone has an idea how to succeed in it, I'm all ears !
> >
> > Thanks,
> >
> > Thomas
>
>

-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener

https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20220809/de1afc9c/attachment-0001.html>