dolfin team mailing list archive

[Johan Hake <johan.hake@xxxxxxxxx>]

It is difficult to advice further as it is not clear what you are doing. But 
remember accessing pure local data structures are best done in scaled to local 
sized data structures. 

If you want to keep a global data structure filled with duplicated data points 
you need to start using MPI communication, which might not be what you want, 
and might qualify for a new question thread.

Johan



On Friday October 15 2010 04:10:26 Jack wrote:
> Question #129296 on DOLFIN changed:
> https://answers.launchpad.net/dolfin/+question/129296
> 
> Jack requested for more information:
> Hi
> 
> What if instances is a std::vector is of size
> 'V.dofmap().global.dimension()' ?
> 
> I'm trying to  split it across the processors but it still keeps
> crashing.
> 
> roam146-22:JJ$ mpirun -n 2 ./test
> Process 0: Number of global vertices: 10000
> Process 0: Number of global cells: 19602
> Process 0: Partitioned mesh, edge cut is 102.
> Process 1: Partitioned mesh, edge cut is 102.
> Process 0: Solving linear system of size 10000 x 10000 (PETSc Krylov
> solver). Process 0: PETSc Krylov solver (cg, bjacobi) converged in 7
> iterations. Process 0: PETSc Krylov solver preconditioner (bjacobi)
> sub-methods: (preonly, ilu) Process 1: PETSc Krylov solver (cg, bjacobi)
> converged in 7 iterations. Process 1: PETSc Krylov solver preconditioner
> (bjacobi) sub-methods: (preonly, ilu) Assertion failed: (i < _size),
> function operator[], file /Users/JJ/FEniCS/include/dolfin/common/Array.h,
> line 133.
> 
> Many thanks
> 
> Jack