[petsc-dev] petsc-dev: ASM with sbaij matrix and block size=1 broken?

[Matthew Knepley]

On Sat, Nov 5, 2011 at 12:10 AM, Brad Aagaard <baagaard at usgs.gov> wrote:

> Jed,
>
> It looks like your changeset 20985 may have introduced a bug. I just
> pulled and rebuilt petsc-dev. I am trying to run a PyLith simulation that
> worked in parallel with ASM, sbaij matrix, and bs=1 a week ago but now
> fails. I get a memory allocation error (huge allocation) or a memory
> corruption error. Using an aij matrix or running in serial works fine. My
> simulation runs fine with 2 processes but not 3 or 4. The stack traces are
>

I am trying to understand, but having problems.

1) 20985 does not change anything here. Do you mean 20989?

2) The changes that 20989 makes are only to multiply in two place by
C->rmap->bs. Now maybe that is screwed up.

3) I can't see memory allocation anywhere near here:

line 929
            if ((ctmp = bmap[cworkB[l]]) < cstart)  cols[l] = ctmp;
line 933
          for (l=0; l<nzA; l++)   cols[imark+l] = cstart + cworkA[l];

  Thanks,

     Matt


> NPROC=4 (memory corruption)
> #0  0x00007f014a93bb00 in MatGetSubMatrices_MPIBAIJ_**local (C=0x29560a0,
>    ismax=1, isrow=0x2a06520, iscol=0x29ff6f0, scall=MAT_INITIAL_MATRIX,
>    allrows=0x2a05db0, allcolumns=0x2a00ea0, submats=0x2a0ab00)
>    at /tools/common/petsc-dev/src/**mat/impls/baij/mpi/baijov.c:**933
> #1  0x00007f014aacd73d in MatIncreaseOverlap_MPISBAIJ (C=0x29560a0,
> is_max=1,
>    is=0x29fb110, ov=1)
>    at /tools/common/petsc-dev/src/**mat/impls/sbaij/mpi/sbaijov.c:**85
> #2  0x00007f014ac476db in MatIncreaseOverlap (mat=0x29560a0, n=1,
>    is=0x29fb110, ov=1)
>    at /tools/common/petsc-dev/src/**mat/interface/matrix.c:6669
> #3  0x00007f014b016df6 in PCSetUp_ASM (pc=0x299d2f0)
>    at /tools/common/petsc-dev/src/**ksp/pc/impls/asm/asm.c:199
>
> NPROC=3 (request for huge amount of memory allocation)
> #0  0x00007f67d30bca75 in MatGetSubMatrices_MPIBAIJ_**local (C=0x2cc8430,
>    ismax=1, isrow=0x2eb4b70, iscol=0x2eadaf0, scall=MAT_INITIAL_MATRIX,
>    allrows=0x2eb4400, allcolumns=0x2eaf2a0, submats=0x2eb9150)
>    at /tools/common/petsc-dev/src/**mat/impls/baij/mpi/baijov.c:**929
> #1  0x00007f67d324e73d in MatIncreaseOverlap_MPISBAIJ (C=0x2cc8430,
> is_max=1,
>    is=0x2ea9510, ov=1)
>    at /tools/common/petsc-dev/src/**mat/impls/sbaij/mpi/sbaijov.c:**85
> #2  0x00007f67d33c86db in MatIncreaseOverlap (mat=0x2cc8430, n=1,
>    is=0x2ea9510, ov=1)
>    at /tools/common/petsc-dev/src/**mat/interface/matrix.c:6669
> #3  0x00007f67d3797df6 in PCSetUp_ASM (pc=0x2d0cdf0)
>    at /tools/common/petsc-dev/src/**ksp/pc/impls/asm/asm.c:199
> #4  0x00007f67d36c43a5 in PCSetUp (pc=0x2d0cdf0)
>
> Thanks,
> Brad
>



-- 
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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20111105/40115768/attachment.html>