[petsc-dev] corrupted ta->count with GAMG

Thu Oct 25 12:49:48 CDT 2012

Yes, it looks like you need to configure with this option.

I would be curious about what this option does but I'm not super optimistic that it will help.  I worked on getting GAMG and ML to match and the parameter that is defaulted differently is the threshold value.  I don't have the parameter in front of me but if you run with -help, and ML, then you will get an output of parameters, and you should be able search for "threshold".  ML default to 0.0 and GAMG defaults to about 0.05.  So I would start by fiddling with this parameter to try to match the convergence rates.

Mark

On Oct 25, 2012, at 9:37 AM, John Mousel <john.mousel at gmail.com> wrote:

> Mark,
> 
> I'll try to get you some numbers on this. One quick question. Is it possible to use 
> 
> -pc_ml_CoarsenScheme METIS 
> 
> I've tried, but I get an error message.
> 
> *ML*WRN* This function has been compiled without the configure
> *ML*WRN* option --with-ml_metis
> *ML*WRN* I will put all the nodes in the same aggregate, this time...
> *ML*WRN* (file ./Coarsen/ml_agg_METIS.c, line 966)
> 
> I'm about to build, and I didn't know if I'm missing an option that I need. 
> 
> John
> 
> On Thu, Oct 25, 2012 at 10:24 AM, Mark F. Adams <mark.adams at columbia.edu> wrote:
> 
> On Oct 25, 2012, at 7:30 AM, John Mousel <john.mousel at gmail.com> wrote:
> 
>> That was my first thought as well. I was hoping the changes he made would have brought down GAMG's setup cost some as right now GAMG isn't competetive with ML because my matrix numbering changes every time step.
> 
> Yes, our setup times need work.
> 
>> I get much better performance from GAMG after the setup phase...unfortunately that does me no good!
> 
> ML can get the same performance as GAMG, they are very similar algorithms.  Are your convergence rates better or is the timing just better?
> 
> 
>> 
>> John
>> 
>> On Thu, Oct 25, 2012 at 9:23 AM, Mark F. Adams <mark.adams at columbia.edu> wrote:
>> The problem is detected in MatTranspose and Jed was messing with this a few weeks ago ...
>> 
>> 
>> On Oct 25, 2012, at 6:48 AM, John Mousel <john.mousel at gmail.com> wrote:
>> 
>>> Mark,
>>> 
>>> Yes, this worked before pulling. I hadn't pulled on that machine in a few weeks though.
>>> 
>>> John
>>> 
>>> On Thu, Oct 25, 2012 at 1:18 AM, Mark F. Adams <mark.adams at columbia.edu> wrote:
>>> Did this work before the pull?
>>> 
>>> Mark
>>> 
>>> On Oct 24, 2012, at 8:04 PM, John Mousel <john.mousel at gmail.com> wrote:
>>> 
>>> > I just pulled and I'm getting the error below when running with GAMG.
>>> >
>>> > [20]PETSC ERROR: --------------------- Error Message ------------------------------------
>>> > [20]PETSC ERROR: Corrupted Petsc object!
>>> > [20]PETSC ERROR: corrupted ta->count!
>>> > [20]PETSC ERROR: ------------------------------------------------------------------------
>>> > [20]PETSC ERROR: Petsc Development HG revision: 2b3cccd7bd8ff536f48616e4402b164ba9cd12d0  HG Date: Wed Oct 24 21:15:30 2012 -0500
>>> > [20]PETSC ERROR: See docs/changes/index.html for recent updates.
>>> > [73]PETSC ERROR: [20]PETSC ERROR: --------------------- Error Message ------------------------------------
>>> > See docs/faq.html for hints about trouble shooting.
>>> > [73]PETSC ERROR: [20]PETSC ERROR: See docs/index.html for manual pages.
>>> > Corrupted Petsc object!
>>> > [20]PETSC ERROR: [73]PETSC ERROR: ------------------------------------------------------------------------
>>> > corrupted ta->count!
>>> > [253]PETSC ERROR: --------------------- Error Message ------------------------------------
>>> > [253]PETSC ERROR: Corrupted Petsc object!
>>> > [253]PETSC ERROR: corrupted ta->count!
>>> > [253]PETSC ERROR: ------------------------------------------------------------------------
>>> > [253]PETSC ERROR: Petsc Development HG revision: 2b3cccd7bd8ff536f48616e4402b164ba9cd12d0  HG Date: Wed Oct 24 21:15:30 2012 -0500
>>> > [253]PETSC ERROR: See docs/changes/index.html for recent updates.
>>> > [253]PETSC ERROR: See docs/faq.html for hints about trouble shooting.
>>> > [253]PETSC ERROR: See docs/index.html for manual pages.
>>> > [253]PETSC ERROR: ------------------------------------------------------------------------
>>> > [253]PETSC ERROR: /nfsscratch/Users/jmousel/MHV/MHV_exe on a intel-opt named compute-6-165.local by jmousel Wed Oct 24 21:59:50 2012
>>> > [253]PETSC ERROR: Libraries linked from /Users/jmousel/SOFT/petsc-dev/intel-opt/lib
>>> > [253]PETSC ERROR: Configure run at Wed Oct 24 21:41:09 2012
>>> > [253]PETSC ERROR: Configure options --download-blacs=1 --download-cmake=1 --download-hypre=1 --download-metis=1 --download-ml=1 --download-parmetis=1 --download-scalapack=1 --with-blas-lapack-dir=/opt/intel/mkl/10.2.5.035/lib/em64t --with-debugging=0 --with-mpi-dir=/opt/openmpi/1.4.3/intel COPTFLAGS=-O2 FOPTFLAGS=-O2 PETSC_ARCH=intel-opt
>>> > [253]PETSC ERROR: ------------------------------------------------------------------------
>>> > [253]PETSC ERROR: PetscTableAddExpand() line 145 in src/sys/utils/ctable.c
>>> > [253]PETSC ERROR: PetscTableAdd() line 75 in src/sys/sf//Users/jmousel/SOFT/petsc-dev/include/petscctable.h
>>> > [253]PETSC ERROR: PetscSFSetGraph() line 327 in src/sys/sf/sf.c
>>> > [253]PETSC ERROR: PetscSFSetGraphLayout() line 595 in src/vec/vec/impls/mpi/pmap.c
>>> > [253]PETSC ERROR: MatTranspose_MPIAIJ() line 2130 in src/mat/impls/aij/mpi/mpiaij.c
>>> > [253]PETSC ERROR: MatTranspose() line 4350 in src/mat/interface/matrix.c
>>> > [253]PETSC ERROR: PCGAMGFilterGraph() line 136 in src/ksp/pc/impls/gamg/tools.c
>>> > [253]PETSC ERROR: PCGAMGgraph_AGG() line 979 in src/ksp/pc/impls/gamg/agg.c
>>> > [253]PETSC ERROR: PCSetUp_GAMG() line 685 in src/ksp/pc/impls/gamg/gamg.c
>>> > [253]PETSC ERROR: PCSetUp() line 832 in src/ksp/pc/interface/precon.c
>>> > [253]PETSC ERROR: KSPSetUp() line 278 in src/ksp/ksp/interface/itfunc.c
>>> > [253]PETSC ERROR: KSPSolve() line 402 in src/ksp/ksp/interface/itfunc.c
>>> > [73]PETSC ERROR: [20]PETSC ERROR: ------------------------------------------------------------------------
>>> > /nfsscratch/Users/jmousel/MHV/MHV_exe on a intel-opt named compute-1-16.local by jmousel Wed Oct 24 21:59:50 2012
>>> > [73]PETSC ERROR: [20]PETSC ERROR: Libraries linked from /Users/jmousel/SOFT/petsc-dev/intel-opt/lib
>>> > [35]PETSC ERROR: Petsc Development HG revision: 2b3cccd7bd8ff536f48616e4402b164ba9cd12d0  HG Date: Wed Oct 24 21:15:30 2012 -0500
>>> > [20]PETSC ERROR: [73]PETSC ERROR: See docs/changes/index.html for recent updates.
>>> > Configure run at Wed Oct 24 21:41:09 2012
>>> > --------------------- Error Message ------------------------------------
>>> > [73]PETSC ERROR: [20]PETSC ERROR: [35]PETSC ERROR: Configure options --download-blacs=1 --download-cmake=1 --download-hypre=1 --download-metis=1 --download-ml=1 --download-parmetis=1 --download-scalapack=1 --with-blas-lapack-dir=/opt/intel/mkl/10.2.5.035/lib/em64t --with-debugging=0 --with-mpi-dir=/opt/openmpi/1.4.3/intel COPTFLAGS=-O2 FOPTFLAGS=-O2 PETSC_ARCH=intel-opt
>>> > See docs/faq.html for hints about trouble shooting.
>>> 
>>> 
>> 
>> 
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20121025/379ad302/attachment.html>