February 2012 - petsc-dev - mailman3.cels.anl.gov

tests/benchmarks
by Gerard Gorman 21 Feb '12

21 Feb '12

Hi I would like to harvest a selection of typical use cases for benchmarking PETSc/OpenMP. Ideally they would have features such as preallocating matrices rather than adding one at a time, be easy to configure for different problem sizes etc. Has anyone already put together such a list - what would be considered good/best practice here? Cheers Gerard

1 0

--download-mpich --download-openmpi and mpicc
by Barry Smith 21 Feb '12

21 Feb '12

Matt and Satish, --download-mpich and --download-openmpi should be immediately rejected out of hand if the compilers start with mpi. Please add this support, it is absurd we have to process email issues regarding this. Barry

1 0

Re: [petsc-dev] Problem with MUMPS+metis
by John Fettig 20 Feb '12

20 Feb '12

On Mon, Feb 20, 2012 at 3:21 PM, Sean Farley <sean(a)mcs.anl.gov> wrote: > This hangs for me in the METIS ordering stage >> (src/ksp/ksp/examples/tutorials/ex2): >> >> mpirun -np 1 ./ex2 -n 100 -m 100 -ksp_type preonly -pc_type lu >> -pc_factor_mat_solver_package mumps -mat_mumps_icntl_7 5 -info > > > Thanks, I'll look into this. > It seems to be a problem with metis' interpretation of the vertex weights. If you set the vwgt=0x0 in a debugger when stopped at metis_nodend_ it completes successfully. Also, if you set the weights to 1 inside dmumps_part2.F like: DO I=1,NCMP FILS(I) = 1 ENDDO it also runs fine. I think having zero vertex weights is probably wrong anyways. Regards, John

1 0

Re: [petsc-dev] Problem with MUMPS+metis
by Sean Farley 20 Feb '12

20 Feb '12

> > This hangs for me in the METIS ordering stage > (src/ksp/ksp/examples/tutorials/ex2): > > mpirun -np 1 ./ex2 -n 100 -m 100 -ksp_type preonly -pc_type lu > -pc_factor_mat_solver_package mumps -mat_mumps_icntl_7 5 -info Thanks, I'll look into this.

1 0

Re: [petsc-dev] Problem with MUMPS+metis
by John Fettig 20 Feb '12

20 Feb '12

On Mon, Feb 20, 2012 at 3:10 PM, Sean Farley <sean(a)mcs.anl.gov> wrote: > Yes, I use this one. >> I do specify serial partitioning. Basically all parameters are the same, >> but mumps from petsc just hangs. > > > It would really help me out if you could get a petsc example to hang as > well and tell me what options you ran with so I could try to debug locally. > This hangs for me in the METIS ordering stage (src/ksp/ksp/examples/tutorials/ex2): mpirun -np 1 ./ex2 -n 100 -m 100 -ksp_type preonly -pc_type lu -pc_factor_mat_solver_package mumps -mat_mumps_icntl_7 5 -info John

1 0

Re: [petsc-dev] Problem with MUMPS+metis
by Sean Farley 20 Feb '12

20 Feb '12

> > Yes, I use this one. > I do specify serial partitioning. Basically all parameters are the same, > but mumps from petsc just hangs. It would really help me out if you could get a petsc example to hang as well and tell me what options you ran with so I could try to debug locally.

1 0

Re: [petsc-dev] Problem with MUMPS+metis
by agrayver＠gfz-potsdam.de 20 Feb '12

20 Feb '12

Yes, I use this one. I do specify serial partitioning. Basically all parameters are the same, but mumps from petsc just hangs. I work with ldlt factorization. ----- Reply message ----- From: "Sean Farley" <sean(a)mcs.anl.gov> To: "For users of the development version of PETSc" <petsc-dev(a)mcs.anl.gov> Subject: [petsc-dev] Problem with MUMPS+metis Date: Mon, Feb 20, 2012 20:42 Which version are you using? The only version of metis supported officially by mumps is metis 4.0.3 which contains METIS_WND and was removed in favor of more arguments to METIS_ND. See my patch submitted here: https://listes.ens-lyon.fr/sympa/arc/mumps-users/2011-11/msg00027.html Search on their mailing list indicates others were successful in linking serially. Did you specify the option for mumps to use serial partitioning (through petsc options), out of curiosity?

1 0

Re: [petsc-dev] Problem with MUMPS+metis
by Sean Farley 20 Feb '12

20 Feb '12

> > Which version are you using? The only version of metis supported officially by mumps is metis 4.0.3 which contains METIS_WND and was removed in favor of more arguments to METIS_ND. See my patch submitted here: https://listes.ens-lyon.fr/sympa/arc/mumps-users/2011-11/msg00027.html Search on their mailing list indicates others were successful in linking serially. Did you specify the option for mumps to use serial partitioning (through petsc options), out of curiosity?

1 0

Re: [petsc-dev] Problem with MUMPS+metis
by John Fettig 20 Feb '12

20 Feb '12

Which version are you using? Thanks, John On Mon, Feb 20, 2012 at 2:32 PM, agrayver(a)gfz-potsdam.de < agrayver(a)gfz-potsdam.de> wrote: > This metis is not the newest one. > > Regards, > Alexander > > ----- Reply message ----- > From: "Sean Farley" <sean(a)mcs.anl.gov> > To: "For users of the development version of PETSc" <petsc-dev(a)mcs.anl.gov > > > Subject: [petsc-dev] Problem with MUMPS+metis > Date: Mon, Feb 20, 2012 20:11 > > > I can confirm that. The same problem happens to me, even though with mumps >> and metis obtained from official site this doesn't happen. > > > Hmmm, which versions are you referring to for mumps+metis from the > official site? Last time I checked, mumps couldn't link to the newest > metis. >

1 0

Re: [petsc-dev] Problem with MUMPS+metis
by agrayver＠gfz-potsdam.de 20 Feb '12

20 Feb '12

This metis is not the newest one. Regards, Alexander ----- Reply message ----- From: "Sean Farley" <sean(a)mcs.anl.gov> To: "For users of the development version of PETSc" <petsc-dev(a)mcs.anl.gov> Subject: [petsc-dev] Problem with MUMPS+metis Date: Mon, Feb 20, 2012 20:11 I can confirm that. The same problem happens to me, even though with mumps and metis obtained from official site this doesn't happen. Hmmm, which versions are you referring to for mumps+metis from the official site? Last time I checked, mumps couldn't link to the newest metis.

1 0