[petsc-users] error in calling VecGetArrayf90()
Ethan Coon
ecoon at lanl.gov
Wed Jan 5 13:52:20 CST 2011
See the below program and output (done with dev, but I don't think this
has changed). In F90, the local arrays, gotten by VecGetArrayF90, are
all indexed:
xx_v(1:(Iend-Istart)).
This should really be moot -- It's way easier to follow the example code
for ex50f90.F and pass their arrays into a subroutine which casts the
shape in the "PETSc way", indexed from (Istart:Iend-1).
Ethan
---
#define PETSC_USE_FORTRAN_MODULES 1
#include "finclude/petscsysdef.h"
#include "finclude/petscvecdef.h"
program test
use petsc
implicit none
Vec x
PetscInt Istart, Iend
PetscScalar, pointer:: xx_v(:)
PetscInt i
PetscErrorCode ierr
PetscInt twenty
PetscInt rank
twenty = 20
call PetscInitialize(PETSC_NULL_CHARACTER, ierr)
call mpi_comm_rank(PETSC_COMM_WORLD,rank,ierr)
call VecCreateMPI(PETSC_COMM_WORLD, PETSC_DECIDE, twenty, x, ierr)
call VecGetArrayF90(x, xx_v, ierr)
xx_v = rank+1
call VecRestoreArrayF90(x, xx_v, ierr)
call VecView(x, PETSC_VIEWER_STDOUT_WORLD, ierr)
call VecGetOwnershipRange(x, Istart, Iend)
write (*,*) 'range on rank', rank, ':', Istart, Iend
call VecGetArrayF90(x, xx_v, ierr)
CHKMEMQ
do i = 1,(Iend-Istart)
write (*,*) i, xx_v(i), rank
enddo
CHKMEMQ
call VecRestoreArrayF90(x, xx_v, ierr)
call VecDestroy(x, ierr)
call PetscFinalize(ierr)
end program test
-----
$> ${PETSC_DIR}/${PETSC_ARCH}/bin/mpiexec -n 2 ./test
Vector Object:
type: mpi
Process [0]
1
1
1
1
1
1
1
1
1
1
Process [1]
2
2
2
2
2
2
2
2
2
2
range on rank 1 : 10 20
1 2.0000000000000000 1
2 2.0000000000000000 1
3 2.0000000000000000 1
4 2.0000000000000000 1
5 2.0000000000000000 1
6 2.0000000000000000 1
7 2.0000000000000000 1
8 2.0000000000000000 1
9 2.0000000000000000 1
10 2.0000000000000000 1
range on rank 0 : 0 10
1 1.00000000000000000 0
2 1.00000000000000000 0
3 1.00000000000000000 0
4 1.00000000000000000 0
5 1.00000000000000000 0
6 1.00000000000000000 0
7 1.00000000000000000 0
8 1.00000000000000000 0
9 1.00000000000000000 0
10 1.00000000000000000 0
---
On Wed, 2011-01-05 at 13:21 -0600, Barry Smith wrote:
> I don't think this is correct. You are suppose to use the local indexing for each process. With the strange index starting at 1 instead of 0.
>
>
> Barry
>
>
> On Jan 5, 2011, at 1:03 PM, Ethan Coon wrote:
>
> > On all processors, the array you get is indexed:
> >
> > xx_v(1:(Iend-Istart))
> >
> > while the Istart and Iend are global indices into the global Vec.
> >
> > It's only by luck that the values on proc 3 are correct (this code
> > should probably seg fault as it is accessing memory outside of the
> > bounds of xx_v).
> >
> > To access the array, you'll want:
> >
> > do i=1, (Iend-Istart)
> > write(6,*)'check xx_v',i,xx_v(i),myid
> > enddo
> >
> > or (better yet) pass it into a subroutine to get the array indexed
> > correctly, like demonstrated in the example.
> >
> > Ethan
> >
> > On Wed, 2011-01-05 at 11:23 -0600, Peter Wang wrote:
> >> Thanks, Satish,
> >>
> >> The index of the array is modified to i+1:
> >> !===================
> >> do i=Istart,Iend-1
> >> write(6,*)'check xx_v',i+1,xx_v(i+1),myid
> >> enddo
> >> !===================
> >>
> >> However, only the elements on root process (process 0) and the
> >> last process (process 3) are corrent, is there any ohter logical
> >> error?
> >>
> >> check xx_v 1 3999.9999999999982 0
> >> check xx_v 2 3999.9999999999982 0
> >> check xx_v 3 3999.9999999999982 0
> >> check xx_v 4 3999.9999999999982 0
> >> check xx_v 5 3999.9999999999982 0
> >> check xx_v 6 3000.0000000000005 0
> >> check xx_v 7 3000.0000000000005 0
> >> check xx_v 8 2.61360726650019422E-321 1
> >> check xx_v 9 7.90505033345994471E-323 1
> >> check xx_v 10 1.69759663277221785E-312 1
> >> check xx_v 11 6.16840020108069212E-317 1
> >> check xx_v 12 6.16840316547456717E-317 1
> >> check xx_v 13 6.16832658529946177E-317 1
> >> check xx_v 14 1.99665037664579820E-314 2
> >> check xx_v 15 6.19784009071943448E-317 2
> >> check xx_v 16 6.20249221284067566E-317 2
> >> check xx_v 17 6.20218737433719161E-317 2
> >> check xx_v 18 6.18236051996958238E-317 2
> >> check xx_v 19 6.16840316547456717E-317 2
> >> check xx_v 20 6.18107199676522841E-317 3
> >> check xx_v 21 0.0000000000000000 3
> >> check xx_v 22 0.0000000000000000 3
> >> check xx_v 23 0.0000000000000000 3
> >> check xx_v 24 0.0000000000000000 3
> >> check xx_v 25 0.0000000000000000 3
> >>
> >>
> >>
> >>> Date: Tue, 4 Jan 2011 22:49:50 -0600
> >>> From: balay at mcs.anl.gov
> >>> To: petsc-users at mcs.anl.gov
> >>> Subject: Re: [petsc-users] error in calling VecGetArrayf90()
> >>>
> >>> The global index starts at Istart - but the array index starts at 1
> >> [for fortran array]
> >>>
> >>> Satish
> >>>
> >>> On Tue, 4 Jan 2011, Peter Wang wrote:
> >>>
> >>>>
> >>>> In last question, the pointer xx_v is local data. However, if
> >> write them to the monitor or assign them to another array, the value
> >> is incorrect.
> >>>>
> >>>> The protion of the code to display them on the monitor is like as
> >> following:
> >>>> call MatGetOwnershipRange(A,Istart,Iend,ierr)
> >>>> call VecGetArrayF90(x,xx_v,ierr) ! Vector x is matched with Matrix
> >> A in the same communicator
> >>>>
> >>>> write(*,*)xx_v,myid ! write the poiner array together
> >>>>
> >>>> do i=Istart,Iend-1
> >>>> write(6,*)'check xx_v',i,xx_v(i),myid !write the element of the
> >> array one by one with local range (Istart to Iend-1)
> >>>> enddo
> >>>>
> >>>>
> >>>> =========The result is as following: ( the values of the elements
> >> from 7 to 20 are not correct !!)
> >>>>
> >>>> 3999.9999999999982 3999.9999999999982 3999.9999999999982
> >> 3999.9999999999982 3999.9999999999982 3000.0000000000005
> >> 3000.0000000000005 0
> >>>>
> >>>> 3000.0000000000009 3000.0000000000009 3000.0000000000009
> >> 2000.0000000000011 2000.0000000000011 2000.0000000000000 1
> >>>>
> >>>> 2000.0000000000009 2000.0000000000009 1000.0000000000003
> >> 1000.0000000000003 1000.0000000000003 999.99999999999989 2
> >>>>
> >>>> 1000.0000000000000 0.0000000000000000 0.0000000000000000
> >> 0.0000000000000000 0.0000000000000000 0.0000000000000000 3
> >>>>
> >>>>
> >>>> check xx_v 0 0.0000000000000000 0.0000000000000000 0
> >>>> check xx_v 1 3999.9999999999982 3999.9999999999982 0
> >>>> check xx_v 2 3999.9999999999982 3999.9999999999982 0
> >>>> check xx_v 3 3999.9999999999982 3999.9999999999982 0
> >>>> check xx_v 4 3999.9999999999982 3999.9999999999982 0
> >>>> check xx_v 5 3999.9999999999982 3999.9999999999982 0
> >>>> check xx_v 6 3000.0000000000005 3000.0000000000005 0
> >>>> check xx_v 7 1.99665037664579820E-314 1.99665037664579820E-314 1
> >>>> check xx_v 8 2.61360726650019422E-321 2.61360726650019422E-321 1
> >>>> check xx_v 9 7.90505033345994471E-323 7.90505033345994471E-323 1
> >>>> check xx_v 10 1.69759663277221785E-312 1.69759663277221785E-312 1
> >>>> check xx_v 11 6.16846344148335980E-317 6.16846344148335980E-317 1
> >>>> check xx_v 12 6.16846640587723485E-317 6.16846640587723485E-317 1
> >>>> check xx_v 13 6.16838982570212945E-317 6.16838982570212945E-317 2
> >>>> check xx_v 14 1.99665037664579820E-314 1.99665037664579820E-314 2
> >>>> check xx_v 15 6.19790333112210216E-317 6.19790333112210216E-317 2
> >>>> check xx_v 16 6.20255545324334334E-317 6.20255545324334334E-317 2
> >>>> check xx_v 17 6.20225061473985929E-317 6.20225061473985929E-317 2
> >>>> check xx_v 18 6.18242376037225006E-317 6.18242376037225006E-317 2
> >>>> check xx_v 19 6.16846640587723485E-317 6.16846640587723485E-317 3
> >>>> check xx_v 20 6.18113523716789609E-317 6.18113523716789609E-317 3
> >>>> check xx_v 21 0.0000000000000000 0.0000000000000000 3
> >>>> check xx_v 22 0.0000000000000000 0.0000000000000000 3
> >>>> check xx_v 23 0.0000000000000000 0.0000000000000000 3
> >>>> check xx_v 24 0.0000000000000000 0.0000000000000000 3
> >>>>
> >>>> ======The vector x is :
> >>>> Process [0]
> >>>> 4000
> >>>> 4000
> >>>> 4000
> >>>> 4000
> >>>> 4000
> >>>> 3000
> >>>> 3000
> >>>> Process [1]
> >>>> 3000
> >>>> 3000
> >>>> 3000
> >>>> 2000
> >>>> 2000
> >>>> 2000
> >>>> Process [2]
> >>>> 2000
> >>>> 2000
> >>>> 1000
> >>>> 1000
> >>>> 1000
> >>>> 1000
> >>>> Process [3]
> >>>> 1000
> >>>> 0
> >>>> 0
> >>>> 0
> >>>> 0
> >>>> 0
> >>>>
> >>>>
> >>>>
> >>>>
> >>>>> Date: Tue, 4 Jan 2011 17:50:11 -0600
> >>>>> From: balay at mcs.anl.gov
> >>>>> To: petsc-users at mcs.anl.gov
> >>>>> Subject: Re: [petsc-users] error in calling VecGetArrayf90()
> >>>>>
> >>>>> Did you included "finclude/petscvec.h90" in your code - as the
> >> example did?
> >>>>>
> >>>>> satish
> >>>>>
> >>>>> On Tue, 4 Jan 2011, Peter Wang wrote:
> >>>>>
> >>>>>>
> >>>>>> I am trying to obtain the value of each element of a solution
> >> Vector by KSPsolve().
> >>>>>>
> >>>>>> The variables are defined according the example of ex4f90.F in
> >> \petsc-3.1-p5\src\snes\examples\tutorials\ as following,
> >>>>>>
> >>>>>> PetscScalar, pointer :: xx_v(:)
> >>>>>>
> >>>>>> ...
> >>>>>> call KSPSolve(ksp,b,x,ierr)
> >>>>>> call VecView(x,PETSC_VIEWER_STDOUT_WORLD,ierr)
> >>>>>>
> >>>>>> call VecGetArrayF90(x,xx_v,ierr)
> >>>>>> call VecRestoreArrayF90(x,xx_v,ierr)
> >>>>>>
> >>>>>> ...
> >>>>>>
> >>>>>> But, the error keeps coming out when call
> >> VecGetArrayF90(x,xx_v,ierr) and call VecRestoreArrayF90(x,xx_v,ierr)
> >> are not commented off.
> >>>>>>
> >>>>>>
> >>>>>> The error information shows:
> >>>>>> Caught signal number 11 SEGV: Segmentation Violation, probably
> >> memory access out of range
> >>>>>>
> >>>>>> [0]PETSC ERROR: --------------------- Stack Frames
> >> ------------------------------------
> >>>>>> [0]PETSC ERROR: Note: The EXACT line numbers in the stack are
> >> not available,
> >>>>>> [0]PETSC ERROR: INSTEAD the line number of the start of the
> >> function
> >>>>>> [0]PETSC ERROR: is given.
> >>>>>> [0]PETSC ERROR: [0] F90Array1dCreate line 52
> >> src/sys/f90-src/f90_cwrap.c
> >>>>>> [0]PETSC ERROR: --------------------- Error Message
> >> ------------------------------------
> >>>>>>
> >>>>>> I checked the code according the example, but cannot see any
> >> difference to that. Just don't know why the pointer array xx_v doesn't
> >> work here? Thanks.
> >>>>>>
> >>>>>>
> >>>>>
> >>>>
> >>>
> >
> > --
> > -------------------------------------
> > Ethan Coon
> > Post-Doctoral Researcher
> > Mathematical Modeling and Analysis
> > Los Alamos National Laboratory
> > 505-665-8289
> >
> > http://www.ldeo.columbia.edu/~ecoon/
> > -------------------------------------
> >
>
--
-------------------------------------
Ethan Coon
Post-Doctoral Researcher
Mathematical Modeling and Analysis
Los Alamos National Laboratory
505-665-8289
http://www.ldeo.columbia.edu/~ecoon/
-------------------------------------
More information about the petsc-users
mailing list