On Tue, Mar 3, 2009 at 7:54 AM, STEPHANE TCHOUANMO <tchouanm at msn.com> wrote:
> Hi all, > > thank you Barry for the indication you gave me. > > As a matter of fact, i verified my jacobian and function evaluation again > and again but i really dont see anything wrong in it. > So i came back to the basic Laplacian problem (- \Delta u = f ) in the unit > cube discretized in regular hexes. The numerical scheme i use is a > vertex-centred finite volume scheme. > The solution i get is correct compared to the exact solution (of second > order) and i know my jacobian and residual evalutions are correct. But here > is the log out i get. > > > Event Count Time (sec) > Flops/sec --- Global --- --- Stage --- Total > Max Ratio Max Ratio Max Ratio Mess Avg len > Reduct %T %F %M %L %R %T %F %M %L %R Mflop/s > > ------------------------------------------------------------------------------------------------------------------------ > > --- Event Stage 0: Main Stage > > VecMDot 71 1.0 2.9587e-02 1.0 6.23e+08 1.0 0.0e+00 0.0e+00 > 0.0e+00 0 25 0 0 0 0 25 0 0 0 623 > VecNorm 77 1.0 3.3638e-02 1.0 4.24e+07 1.0 0.0e+00 0.0e+00 > 0.0e+00 0 2 0 0 0 0 2 0 0 0 42 > VecScale 74 1.0 2.1052e-03 1.0 3.26e+08 1.0 0.0e+00 0.0e+00 > 0.0e+00 0 1 0 0 0 0 1 0 0 0 326 > VecCopy 80 1.0 3.4863e-03 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 > VecSet 9 1.0 2.0776e-04 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 > VecAXPY 5 1.0 2.3208e-04 1.0 3.99e+08 1.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 399 > VecWAXPY 1 1.0 6.6995e-05 1.0 1.38e+08 1.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 138 > VecMAXPY 74 1.0 3.8138e-02 1.0 5.18e+08 1.0 0.0e+00 0.0e+00 > 0.0e+00 0 27 0 0 0 0 27 0 0 0 518 > VecAssemblyBegin 4 1.0 9.8636e-06 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 > VecAssemblyEnd 4 1.0 6.9494e-06 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 > VecScatterBegin 3 1.0 3.0706e-04 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 > VecNormalize 74 1.0 3.4648e-02 1.0 5.88e+07 1.0 0.0e+00 0.0e+00 > 0.0e+00 0 3 0 0 0 0 3 0 0 0 59 > MatMult 73 1.0 1.4618e-01 1.0 2.22e+08 1.0 0.0e+00 0.0e+00 > 0.0e+00 0 45 0 0 0 0 45 0 0 0 222 > MatAssemblyBegin 2 1.0 6.9899e-06 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 > MatAssemblyEnd 2 1.0 6.1999e-03 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 > SNESSolve 1 1.0 6.7333e+01 1.0 1.08e+06 1.0 0.0e+00 0.0e+00 > 3.0e+00 99100 0 0100 99100 0 0100 1 > SNESLineSearch 1 1.0 5.1989e-01 1.0 8.91e+04 1.0 0.0e+00 0.0e+00 > 1.0e+00 1 0 0 0 33 1 0 0 0 33 0 > SNESFunctionEval 2 1.0 1.0441e+00 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 2.0e+00 2 0 0 0 67 2 0 0 0 67 0 > SNESJacobianEval 1 1.0 6.6026e+01 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 1.0e+00 97 0 0 0 33 97 0 0 0 33 0 > KSPGMRESOrthog 71 1.0 6.5884e-02 1.0 5.60e+08 1.0 0.0e+00 0.0e+00 > 0.0e+00 0 51 0 0 0 0 51 0 0 0 560 > KSPSetup 1 1.0 2.2203e-04 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 > KSPSolve 1 1.0 2.6036e-01 1.0 2.80e+08 1.0 0.0e+00 0.0e+00 > 0.0e+00 0100 0 0 0 0100 0 0 0 280 > PCSetUp 1 1.0 7.9495e-06 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 > PCApply 74 1.0 3.6445e-03 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 > 0.0e+00 0 0 0 0 0 0 0 0 0 0 0 > > ------------------------------------------------------------------------------------------------------------------------ > > Memory usage is given in bytes: > > Object Type Creations Destructions Memory Descendants' Mem. > > --- Event Stage 0: Main Stage > > Index Set 3 3 111792 0 > Vec 44 3 223596 0 > Vec Scatter 3 3 0 0 > Matrix 1 0 0 0 > SNES 1 0 0 0 > Krylov Solver 1 0 0 0 > Preconditioner 1 0 0 0 > Viewer 2 0 0 0 > Draw 1 0 0 0 > > ======================================================================================================================== > Average time to get PetscTime(): 1.60268e-06 > > > This shows that the Jacobian evaluation takes 97% of time and the residual > just 2% in the SNESSolve. But if you look at the total MFlops, you can see > that its null(i guess very low) for these phases. What seems to be long is > the part in red concerning Vector manips. You can even see at the end that > the most memory use is in Index set and Vec. > This analysis does not make sense. If you add all the time spent in the Vec operations (in red), it is less than 1/100 of the time in the SNES Solve. There is obviously a problem in that routine, if there is indeed a problem. Do you have a model of the computation that says that this time is too long? Matt -- 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-users/attachments/20090303/71ecdee5/attachment.htm>
