Manfred :<div>I tested 'SAME_NONZERO_PATTERN' with sequential mumps using</div><div>petsc-dev/src/ksp/ksp/examples/tutorials/ex5.c:</div><div>petsc-dev/src/ksp/ksp/examples/tutorials>./ex5 -ksp_monitor -mat_type sbaij -pc_type cholesky -pc_factor_mat_solver_package mumps</div>
<div><div> 0 KSP Residual norm 7.416198487096e+00 </div><div> 1 KSP Residual norm 8.281965561733e-16 </div><div> 0 KSP Residual norm 7.416198487096e+00 </div><div> 1 KSP Residual norm 9.211962217404e-16 </div></div><div>
<br></div><div>Can you switch to petsc-dev </div><div>(<a href="http://www.mcs.anl.gov/petsc/developers/index.html">http://www.mcs.anl.gov/petsc/developers/index.html</a>) to see if you still get the same error?</div><div>
<br></div><div>Hong</div><div><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><br>
<br>
I am solving with cholesky finite element problems where I have to solve multiple times. To improve the performance I use KSPSetOperators and set the option SAME_NONZERO_PATTERN because the matrix structure is always the same in successive solves. This works great as long I use more than one processor. When I use MUMPS, only one processor and a matrix made with MATSEQSBAIJ the solving segfaults in the second solve. When I use spooles or use LU with AIJ it also works, but I need MUMPS as MatGetInertia only works with MUMPS with multiple processors.<br>
<br>
The segfault:<br>
[0]PETSC ERROR: ------------------------------<u></u>------------------------------<u></u>------------<br>
[0]PETSC ERROR: Caught signal number 11 SEGV: Segmentation Violation, probably memory access out of range<br>
[0]PETSC ERROR: Try option -start_in_debugger or -on_error_attach_debugger<br>
[0]PETSC ERROR: or see <a href="http://www.mcs.anl.gov/petsc/petsc-as/documentation/faq.html#valgrind[0]PETSC" target="_blank">http://www.mcs.anl.gov/petsc/<u></u>petsc-as/documentation/faq.<u></u>html#valgrind[0]PETSC</a> ERROR: or try <a href="http://valgrind.org" target="_blank">http://valgrind.org</a> on GNU/linux and Apple Mac OS X to find memory corruption errors<br>
[0]PETSC ERROR: likely location of problem given in stack below<br>
[0]PETSC ERROR: --------------------- Stack Frames ------------------------------<u></u>------<br>
[0]PETSC ERROR: Note: The EXACT line numbers in the stack are not available,<br>
[0]PETSC ERROR: INSTEAD the line number of the start of the function<br>
[0]PETSC ERROR: is given.<br>
[0]PETSC ERROR: [0] MatFactorNumeric_MUMPS line 642 src/mat/impls/aij/mpi/mumps/<u></u>mumps.c<br>
[0]PETSC ERROR: [0] MatCholeskyFactorNumeric line 3037 src/mat/interface/matrix.c<br>
[0]PETSC ERROR: [0] PCSetUp_Cholesky line 92 src/ksp/pc/impls/factor/<u></u>cholesky/cholesky.c<br>
[0]PETSC ERROR: [0] PCSetUp line 797 src/ksp/pc/interface/precon.c<br>
[0]PETSC ERROR: [0] KSPSetUp line 184 src/ksp/ksp/interface/itfunc.c<br>
[0]PETSC ERROR: [0] KSPSolve line 331 src/ksp/ksp/interface/itfunc.c<br>
[0]PETSC ERROR: --------------------- Error Message ------------------------------<u></u>------<br>
[0]PETSC ERROR: Signal received!<br>
[0]PETSC ERROR: ------------------------------<u></u>------------------------------<u></u>------------<br>
[0]PETSC ERROR: Petsc Release Version 3.2.0, Patch 6, Wed Jan 11 09:28:45 CST 2012<br>
[0]PETSC ERROR: See docs/changes/index.html for recent updates.<br>
[0]PETSC ERROR: See docs/faq.html for hints about trouble shooting.<br>
[0]PETSC ERROR: See docs/index.html for manual pages.<br>
[0]PETSC ERROR: ------------------------------<u></u>------------------------------<u></u>------------<br>
[0]PETSC ERROR: ./icona on a linux-gnu named <a href="http://sunray.dps.uibk.ac.at" target="_blank">sunray.dps.uibk.ac.at</a> by manfred Thu Mar 1 10:06:47 2012<br>
[0]PETSC ERROR: Libraries linked from /home/manfred/petsc-3.2-p6/<u></u>linux-gnu-c/lib<br>
[0]PETSC ERROR: Configure run at Thu Mar 1 09:11:08 2012<br>
[0]PETSC ERROR: Configure options --with-cc=/home/manfred/gcc-4.<u></u>6.2/bin/gcc --download-mpich=1 --with-blas-lapack=1 --download-f-blas-lapack=1 PETSC_ARCH=linux-gnu-c --with-clanguage=cxx --with-cxx=/home/manfred/gcc-<u></u>4.6.2/bin/g++ --with-fc=/home/manfred/gcc-4.<u></u>6.2/bin/gfortran --with-scalapack=1 --download-scalapack=1 --with-blacs=1 --download-blacs=1 --with-parmetis=1 --download-parmetis=1 --with-spooles=1 --download-spooles=1 --with-mumps=1 --download-mumps=1 --with-blopex=1 --download-blopex=1 --with-debugging=1<br>
[0]PETSC ERROR: ------------------------------<u></u>------------------------------<u></u>------------<br>
[0]PETSC ERROR: User provided function() line 0 in unknown directory unknown file<br>
application called MPI_Abort(MPI_COMM_WORLD, 59) - process 0<br>
[cli_0]: aborting job:<br>
application called MPI_Abort(MPI_COMM_WORLD, 59) - process 0<br>
<br>
<br>
with valgrind I get the error<br>
==6853== Invalid read of size 8<br>
==6853== at 0x11CDDAA: dmumps_148_ (dmumps_part1.F:1299)<br>
==6853== by 0x111093C: dmumps_142_ (dmumps_part5.F:5094)<br>
==6853== by 0x11C9754: dmumps_ (dmumps_part1.F:642)<br>
==6853== by 0x10B4DDB: dmumps_f77_ (dmumps_part3.F:6651)<br>
==6853== by 0x108E0FF: dmumps_c (mumps_c.c:422)<br>
==6853== by 0xAE6EB3: MatFactorNumeric_MUMPS(_p_Mat*<u></u>, _p_Mat*, MatFactorInfo const*) (mumps.c:667)<br>
==6853== by 0x9832D7: MatCholeskyFactorNumeric(_p_<u></u>Mat*, _p_Mat*, MatFactorInfo const*) (matrix.c:3050)<br>
==6853== by 0xDFFE2D: PCSetUp_Cholesky(_p_PC*) (cholesky.c:157)<br>
==6853== by 0xD7B390: PCSetUp(_p_PC*) (precon.c:819)<br>
==6853== by 0xE47BD2: KSPSetUp(_p_KSP*) (itfunc.c:260)<br>
==6853== by 0xE48C47: KSPSolve(_p_KSP*, _p_Vec*, _p_Vec*) (itfunc.c:379)<br>
==6853== by 0x4914E0: Task::runCalc() (Task.cpp:1204)<br>
<br>
Is this a bug or doing I am something wrong? Should I go to MUMPS with this error? I have currently a workaround to simply not use SAME_NONZERO_PATTERN if using only one processor but as a result it is much slower in successive solves.<br>
<br>
Thank you in advance<span><font color="#888888"><br>
Manfred Gratt<br>
<br>
</font></span></blockquote></div><br></div>