On Thu, Oct 17, 2013 at 7:43 AM, Bishesh Khanal <[email protected]> wrote:
> > > > On Thu, Oct 17, 2013 at 1:07 PM, Matthew Knepley <[email protected]>wrote: > >> On Thu, Oct 17, 2013 at 3:42 AM, Bishesh Khanal <[email protected]>wrote: >> >>> >>> >>> >>> On Wed, Oct 16, 2013 at 8:04 PM, Satish Balay <[email protected]> wrote: >>> >>>> On Wed, 16 Oct 2013, Matthew Knepley wrote: >>>> >>>> > You can also try running under MPICH, which can be valgrind clean. >>>> >>>> Actually --download-mpich would configure/install mpich with >>>> appropriate flags to be valgrind clean. >>>> >>> >>> In my laptop (but not in the cluster, please look at the second part of >>> this reply below for the cluster case) that's how I configured petsc and >>> ran it under mpich. The following errors (which I do not understand what >>> they mean) was reported by valgrind when using the mpich of the petsc in my >>> laptop: Here is the command I used and the error: >>> >> >> This is harmless, and as you can see it comes from gfortran >> initialization. >> >> >>> (Note: petsc is an alias in my .bashrc: alias >>> petsc='/home/bkhanal/Documents/softwares/petsc-3.4.3/bin/petscmpiexec' >>> >>> petsc -n 2 valgrind src/AdLemMain -pc_type fieldsplit >>> -pc_fieldsplit_type schur -pc_fieldsplit_dm_splits 0 >>> -pc_fieldsplit_0_fields 0,1,2 -pc_fieldsplit_1_fields 3 >>> -fieldsplit_0_pc_type hypre -fieldsplit_0_ksp_converged_reason >>> -ksp_converged_reason >>> ==3106== Memcheck, a memory error detector >>> ==3106== Copyright (C) 2002-2010, and GNU GPL'd, by Julian Seward et al. >>> ==3106== Using Valgrind-3.6.1 and LibVEX; rerun with -h for copyright >>> info >>> ==3107== Memcheck, a memory error detector >>> ==3107== Copyright (C) 2002-2010, and GNU GPL'd, by Julian Seward et al. >>> ==3107== Using Valgrind-3.6.1 and LibVEX; rerun with -h for copyright >>> info >>> ==3107== Command: src/AdLemMain -pc_type fieldsplit -pc_fieldsplit_type >>> schur -pc_fieldsplit_dm_splits 0 -pc_fieldsplit_0_fields 0,1,2 >>> -pc_fieldsplit_1_fields 3 -fieldsplit_0_pc_type hypre >>> -fieldsplit_0_ksp_converged_reason -ksp_converged_reason >>> ==3107== >>> ==3106== Command: src/AdLemMain -pc_type fieldsplit -pc_fieldsplit_type >>> schur -pc_fieldsplit_dm_splits 0 -pc_fieldsplit_0_fields 0,1,2 >>> -pc_fieldsplit_1_fields 3 -fieldsplit_0_pc_type hypre >>> -fieldsplit_0_ksp_converged_reason -ksp_converged_reason >>> ==3106== >>> ==3107== Conditional jump or move depends on uninitialised value(s) >>> ==3107== at 0x32EEED9BCE: ??? (in /usr/lib64/libgfortran.so.3.0.0) >>> ==3107== by 0x32EEED9155: ??? (in /usr/lib64/libgfortran.so.3.0.0) >>> ==3107== by 0x32EEE185D7: ??? (in /usr/lib64/libgfortran.so.3.0.0) >>> ==3107== by 0x32ECC0F195: call_init.part.0 (in /lib64/ld-2.14.90.so) >>> ==3107== by 0x32ECC0F272: _dl_init (in /lib64/ld-2.14.90.so) >>> ==3107== by 0x32ECC01719: ??? (in /lib64/ld-2.14.90.so) >>> ==3107== by 0xE: ??? >>> ==3107== by 0x7FF0003EE: ??? >>> ==3107== by 0x7FF0003FC: ??? >>> ==3107== by 0x7FF000405: ??? >>> ==3107== by 0x7FF000410: ??? >>> ==3107== by 0x7FF000424: ??? >>> ==3107== >>> ==3107== Conditional jump or move depends on uninitialised value(s) >>> ==3107== at 0x32EEED9BD9: ??? (in /usr/lib64/libgfortran.so.3.0.0) >>> ==3107== by 0x32EEED9155: ??? (in /usr/lib64/libgfortran.so.3.0.0) >>> ==3107== by 0x32EEE185D7: ??? (in /usr/lib64/libgfortran.so.3.0.0) >>> ==3107== by 0x32ECC0F195: call_init.part.0 (in /lib64/ld-2.14.90.so) >>> ==3107== by 0x32ECC0F272: _dl_init (in /lib64/ld-2.14.90.so) >>> ==3107== by 0x32ECC01719: ??? (in /lib64/ld-2.14.90.so) >>> ==3107== by 0xE: ??? >>> ==3107== by 0x7FF0003EE: ??? >>> ==3107== by 0x7FF0003FC: ??? >>> ==3107== by 0x7FF000405: ??? >>> ==3107== by 0x7FF000410: ??? >>> ==3107== by 0x7FF000424: ??? >>> ==3107== >>> dmda of size: (8,8,8) >>> >>> using schur complement >>> >>> using user defined split >>> Linear solve converged due to CONVERGED_ATOL iterations 0 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 3 >>> Linear solve converged due to CONVERGED_RTOL iterations 1 >>> ==3106== >>> ==3106== HEAP SUMMARY: >>> ==3106== in use at exit: 187,709 bytes in 1,864 blocks >>> ==3106== total heap usage: 112,891 allocs, 111,027 frees, 19,838,487 >>> bytes allocated >>> ==3106== >>> ==3107== >>> ==3107== HEAP SUMMARY: >>> ==3107== in use at exit: 212,357 bytes in 1,870 blocks >>> ==3107== total heap usage: 112,701 allocs, 110,831 frees, 19,698,341 >>> bytes allocated >>> ==3107== >>> ==3106== LEAK SUMMARY: >>> ==3106== definitely lost: 0 bytes in 0 blocks >>> ==3106== indirectly lost: 0 bytes in 0 blocks >>> ==3106== possibly lost: 0 bytes in 0 blocks >>> ==3106== still reachable: 187,709 bytes in 1,864 blocks >>> ==3106== suppressed: 0 bytes in 0 blocks >>> ==3106== Rerun with --leak-check=full to see details of leaked memory >>> ==3106== >>> ==3106== For counts of detected and suppressed errors, rerun with: -v >>> ==3106== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 2 from 2) >>> ==3107== LEAK SUMMARY: >>> ==3107== definitely lost: 0 bytes in 0 blocks >>> ==3107== indirectly lost: 0 bytes in 0 blocks >>> ==3107== possibly lost: 0 bytes in 0 blocks >>> ==3107== still reachable: 212,357 bytes in 1,870 blocks >>> ==3107== suppressed: 0 bytes in 0 blocks >>> ==3107== Rerun with --leak-check=full to see details of leaked memory >>> ==3107== >>> ==3107== For counts of detected and suppressed errors, rerun with: -v >>> ==3107== Use --track-origins=yes to see where uninitialised values come >>> from >>> ==3107== ERROR SUMMARY: 2 errors from 2 contexts (suppressed: 2 from 2) >>> >>> In the above example, the solver iterates and gives results. >>> >>> Now the case in cluster: I had to configure petsc with the option: >>> --with-mpi-dir=/opt/openmpi-gcc/current/ , that's how the cluster >>> administrators asked me to install to get petsc running in many nodes of >>> the clusters. I had tried on my own to configure with --download-mpich in >>> the cluster too, but could not succeed with some errors. If you really >>> think the errors could be from this configuration, I would retry to install >>> with the petscmpich; please let me know. >>> And the valgrind errors for the case where program terminates without >>> completing normally (big sized domain), it has following errors just before >>> abrupt termination: >>> >>> ... lots of other errors and then warnings such as: >>> >> >> This appears to be a bug in OpenMPI, which would not be all that >> surprising. First, you can try running >> in the debugger and extracting a stack trace from the SEGV. >> > > Just to make sure that I understood what you said before I talk with the > cluster administrators: > The program crashes only for a bigger domain size. Even in the cluster, it > does not crash for the domain size up to a certain size. So I need to run > in the debugger for the case when it crashes to get the stack trace from > the SEGV, right ? I do not know how to attach a debugger when submitting a > job to the cluster if that is possible at all! Or are you asking me to run > the program in the debugger in my laptop for the biggest size ? (I have not > tried running the code for the biggest size in my laptop fearing it might > take forever) > There is no use being afraid. Run it both places. Ask the admins how to map your display for the cluster and use -start_in_debugger. Matt > Then you could >> >> 1) Get the admin to install MPICH >> >> 2) Try running a PETSc example on the cluster >> > >> 3) Try running on another machine >> >> Matt >> >> >>> ==55437== Warning: set address range perms: large range [0xc4369040, >>> 0xd6abb670) (defined) >>> ==55438== Warning: set address range perms: large range [0xc4369040, >>> 0xd6a6cd00) (defined) >>> ==37183== Warning: set address range perms: large range [0xc4369040, >>> 0xd69f57d8) (defined) >>> ==37182== Warning: set address range perms: large range [0xc4369040, >>> 0xd6a474f0) (defined) >>> mpiexec: killing job... >>> >>> >>> In between there are several errors such as: >>> ==59334== Use of uninitialised value of size 8 >>> ==59334== at 0xD5B3704: mca_pml_ob1_send_request_put >>> (pml_ob1_sendreq.c:1217) >>> ==59334== by 0xE1EF01A: btl_openib_handle_incoming >>> (btl_openib_component.c:3092) >>> ==59334== by 0xE1F03E9: btl_openib_component_progress >>> (btl_openib_component.c:3634) >>> ==59334== by 0x81CF16A: opal_progress (opal_progress.c:207) >>> ==59334== by 0x81153AC: ompi_request_default_wait_all (condition.h:92) >>> ==59334== by 0xF4C25DD: ompi_coll_tuned_sendrecv_actual >>> (coll_tuned_util.c:54) >>> ==59334== by 0xF4C91FD: >>> ompi_coll_tuned_allgatherv_intra_neighborexchange (coll_tuned_util.h:57) >>> ==59334== by 0x8121783: PMPI_Allgatherv (pallgatherv.c:139) >>> ==59334== by 0x5156D19: ISAllGather (iscoloring.c:502) >>> ==59334== by 0x57A6B78: MatGetSubMatrix_MPIAIJ (mpiaij.c:3607) >>> ==59334== by 0x532DB36: MatGetSubMatrix (matrix.c:7297) >>> ==59334== by 0x5B97725: PCSetUp_FieldSplit(_p_PC*) (fieldsplit.c:524) >>> >>> >>> >>> >>> >>> >>>> >>>> Satish >>>> >>> >>> >> >> >> -- >> 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 >> > > -- 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
