[petsc-users] ILUTP in PETSc

Wed May 14 11:37:57 CDT 2014

On May 14, 2014, at 9:48 AM, Qin Lu <lu_qin_2000 at yahoo.com> wrote:

> It turns out that I can not set PC side as right when KSP type is set to pre_only. After I fixed that, it works fine.
>  
> This brings me a question: why do I have to set KSP type to pre_only when SuperLU's ILUTP is used as preconditioner?

 You don’t and you shouldn’t.  Since you are using SuperLU ILU as a preconditioner, not a direct solver, using preonly means that it will apply the ILU triangular solves only once which means it will not return the correct solution. For LU you can use preonly since LU is a direct (“exact”) solver.

> Can I still set KSP type as KSPBCGS (which seems to be the fastest with PETSc's ILU for my cases)?

    Yes. 

   Barry

>  
> Thanks,
> Qin
> 
> From: Barry Smith <bsmith at mcs.anl.gov>
> To: Qin Lu <lu_qin_2000 at yahoo.com> 
> Cc: Xiaoye S. Li <xsli at lbl.gov>; "petsc-users at mcs.anl.gov" <petsc-users at mcs.anl.gov> 
> Sent: Tuesday, May 13, 2014 8:31 PM
> Subject: Re: [petsc-users] ILUTP in PETSc
> 
> 
>   Works fine for me.  Please please please ALWAYS cut and paste the entire error message that is printed. We print the information for a reason, because it provides clues as to what went wrong.
> 
> 
>   ./ex10 -f0 ~/Datafiles/Matrices/arco1 -pc_type ilu -pc_factor_mat_solver_package superlu -mat_superlu_ilu_droptol 1.e-8 -ksp_monitor_true_residual -ksp_rtol 1.e-12 -ksp_view
> 
>   0 KSP preconditioned resid norm 2.544968580491e+03 true resid norm 7.410897708964e+00 ||r(i)||/||b|| 1.000000000000e+00
>   1 KSP preconditioned resid norm 2.467110329809e-06 true resid norm 1.439993537311e-07 ||r(i)||/||b|| 1.943075716143e-08
>   2 KSP preconditioned resid norm 1.522204461523e-12 true resid norm 2.699724724531e-11 ||r(i)||/||b|| 3.642911871885e-12
> KSP Object: 1 MPI processes
>   type: gmres
>     GMRES: restart=30, using Classical (unmodified) Gram-Schmidt Orthogonalization with no iterative refinement
>     GMRES: happy breakdown tolerance 1e-30
>   maximum iterations=10000, initial guess is zero
>   tolerances:  relative=1e-12, absolute=1e-50, divergence=10000
>   left preconditioning
>   using PRECONDITIONED norm type for convergence test
> PC Object: 1 MPI processes
>   type: ilu
>     ILU: out-of-place factorization
>     0 levels of fill
>     tolerance for zero pivot 2.22045e-14
>     using diagonal shift on blocks to prevent zero pivot [INBLOCKS]
>     matrix ordering: natural
>     factor fill ratio given 0, needed 0
>       Factored matrix follows:
>         Mat Object:        1 MPI processes
>           type: seqaij
>           rows=1501, cols=1501
>           package used to perform factorization: superlu
>           total: nonzeros=0, allocated nonzeros=0
>           total number of mallocs used during MatSetValues calls =0
>             SuperLU run parameters:
>               Equil: YES
>               ColPerm: 3
>               IterRefine: 0
>               SymmetricMode: NO
>               DiagPivotThresh: 0.1
>               PivotGrowth: NO
>               ConditionNumber: NO
>               RowPerm: 1
>               ReplaceTinyPivot: NO
>               PrintStat: NO
>               lwork: 0
>               ILU_DropTol: 1e-08
>               ILU_FillTol: 0.01
>               ILU_FillFactor: 10
>               ILU_DropRule: 9
>               ILU_Norm: 2
>               ILU_MILU: 0
>   linear system matrix = precond matrix:
>   Mat Object:  1 MPI processes
>     type: seqaij
>     rows=1501, cols=1501
>     total: nonzeros=26131, allocated nonzeros=26131
>     total number of mallocs used during MatSetValues calls =0
>       using I-node routines: found 501 nodes, limit used is 5
> Number of iterations =  2
> Residual norm 2.69972e-11
> ~/Src/petsc/src/ksp/ksp/examples/tutorials  master 
> 
> On May 13, 2014, at 12:17 PM, Qin Lu <lu_qin_2000 at yahoo.com> wrote:
> 
> > I tried to use command line options as the example suggested ('-ksp_type preonly -pc_type ilu -pc_factor_mat_solver_package superlu -mat_superlu_ilu_droptol 1.e-8') without changing my source code, but then the call to KSPSetUp returned error number 56.
> >  
> > Does this mean I still need to change the source code (such as adding calls to PCFactorSetMatSolverPackage, PCFactorGetMatrix, etc.)in addition to the command line options?
> >  
> > I ask this since the use of SuperLU seems to be different from using Hypre, which can be invoked with command line options without changing source code.
> >  
> > Thanks a lot,
> > Qin 
> > 
> > 
> > ----- Original Message -----
> > From: Barry Smith <bsmith at mcs.anl.gov>
> > To: Qin Lu <lu_qin_2000 at yahoo.com>
> > Cc: Xiaoye S. Li <xsli at lbl.gov>; "petsc-users at mcs.anl.gov" <petsc-users at mcs.anl.gov>
> > Sent: Monday, May 12, 2014 5:11 PM
> > Subject: Re: [petsc-users] ILUTP in PETSc
> > 
> > 
> >    See for example: http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/Mat/MATSOLVERSUPERLU.html
> > 
> > 
> > 
> > 
> > 
> > On May 12, 2014, at 4:54 PM, Qin Lu <lu_qin_2000 at yahoo.com> wrote:
> > 
> >> Hello,
> >> 
> >> I have built PETSc with SuperLU, but what are PETSc's command line options to invoke SuperLU's ILUTP preconditioner and to set the dropping tolerance? (-mat_superlu_ilu_droptol for the latter?)
> >>  
> >> Do I need to do some programming in order to call SuperLU's preconditioner, or the command line options would work?  
> >>  
> >> Many thanks,
> >> Qin  
> >> 
> >> 
> >>  From: Xiaoye S. Li <xsli at lbl.gov>
> >> To: Barry Smith <bsmith at mcs.anl.gov> 
> >> Cc: Qin Lu <lu_qin_2000 at yahoo.com>; "petsc-users at mcs.anl.gov" <petsc-users at mcs.anl.gov> 
> >> Sent: Friday, May 2, 2014 3:40 PM
> >> Subject: Re: [petsc-users] ILUTP in PETSc
> >> 
> >> 
> >> 
> >> The sequential SuperLU has ILUTP implementation, not in parallel versions. PETSc already supports the option of using SuperLU, so you should be able to try easily.  
> >> 
> >> In SuperLU distribution:
> >> 
> >>    EXAMPLE/zitersol.c : an example to use GMRES with ILUTP preconditioner (returned from driver SRC/zgsisx.c)
> >> 
> >>    SRC/zgsitrf.c : the actual ILUTP factorization routine
> >> 
> >> 
> >> Sherry Li
> >> 
> >> 
> >> 
> >> On Fri, May 2, 2014 at 12:25 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> >> 
> >> 
> >>> At http://www.mcs.anl.gov/petsc/documentation/linearsolvertable.htmlthere are two listed. ./configure —download-hypre
> >>> 
> >>> mpiexec -n 23 ./yourprogram -pc_type hypre -pc_hypre_type ilupt or euclid
> >>> 
> >>> you can also add -help to see what options are available.
> >>> 
> >>>    Both pretty much suck and I can’t image much reason for using them.
> >>> 
> >>>    Barry
> >>> 
> >>> 
> >>> 
> >>> On May 2, 2014, at 10:27 AM, Qin Lu <lu_qin_2000 at yahoo.com> wrote:
> >>> 
> >>>> Hello,
> >>>> 
> >>>> I am interested in using ILUTP preconditioner with PETSc linear solver. There is an online doc https://fs.hlrs.de/projects/par/par_prog_ws/pdf/petsc_nersc01_short.pdfthatmentioned it is available in PETSc with other packages (page 62-63). Is there any instructions or examples on how to use it?
> >>>> 
> >>>> Many thanks,
> >>>> Qin
> >>> 
> >>>      
> 
> 