[petsc-dev] Integer overflow inside GAMG

[Mark Adams]

Yes Pierre, I have see this.  I've worked on this but I'm not sure where
the fixes are in the pipeline or what version you are using.  And it is
tedious to debug stuff at this scale... N.B. There is a real bug on some
compilers (gnu on Cray as I recall) and there is a fix in the pipe, but
this print stuff is separate.
Mark


On Wed, Dec 11, 2013 at 5:21 AM, Pierre Jolivet <jolivet at ann.jussieu.fr>wrote:

> Hello,
> I'm using PETSc without --with-64-bit-indices, and it looks like
> PCGAMGFilterGraph declares variables nnz0 and nnz1 as PetscInt. Shouldn't
> they be declared as Petsc64bitInt (which would default to PetscInt when
> compiling with --with-64-bit-indices) ?
> Here is a "funny" log of GAMG (the execution is fine nonetheless, as you
> might have guessed, this is just for aesthetic consideration):
>
>         [0]PCSetFromOptions_GAMG threshold set 1.000000e-01
>         [0]PCSetUp_GAMG level 0 N=232830963, n data rows=3, n data cols=6,
> nnz/row (ave)=76, np=4096
>         [0]PCGAMGFilterGraph -49.2487% nnz after filtering, with threshold
> 0.1, -26.7705 nnz ave. (N=77610321)
>
> Thank you,
> Pierre
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20131211/6a29b788/attachment.html>