[petsc-users] DOF in DMDACreate3d and staggered grid

TAY wee-beng zonexo at gmail.com
Mon Jun 18 23:04:43 CDT 2012


On 18/6/2012 11:36 PM, Barry Smith wrote:
> On Jun 18, 2012, at 10:34 PM, TAY wee-beng wrote:
>
>> On 18/6/2012 10:58 PM, Barry Smith wrote:
>>> On Jun 18, 2012, at 9:54 PM, TAY wee-beng wrote:
>>>
>>>> On 18/6/2012 10:36 PM, Barry Smith wrote:
>>>>>    Do PETSc examples run on this system. cd src/snes/examples/tutorials; make ex19 ; then run ex19 on a bunch of nodes of the cluster, does it run correctly? Repeat with ex5f
>>>>>
>>>>>     Barry
>>>> ex19 and ex5f both run w/o problem.
>>>>
>>>> [wtay at hpc12:tutorials]$ mpiexec -np 2 ./ex19
>>>> lid velocity = 0.0625, prandtl # = 1, grashof # = 1
>>>> Number of SNES iterations = 2
>>>> [wtay at hpc12:tutorials]$ mpiexec -np 4 ./ex19
>>>> lid velocity = 0.0625, prandtl # = 1, grashof # = 1
>>>> Number of SNES iterations = 2
>>>>
>>>> [wtay at hpc12:tutorials]$ ./ex5f
>>>> Number of SNES iterations =     4
>>>> [wtay at hpc12:tutorials]$ mpiexec -np 2 ./ex5f
>>>> Number of SNES iterations =     4
>>>> [wtay at hpc12:tutorials]$ mpiexec -np 4 ./ex5f
>>>> Number of SNES iterations =     4
>>>> [wtay at hpc12:tutorials]$
>>>>
>>>> The problem seems to lie in DMDAVecGetArrayF90 and DMDAVecRestoreArrayF90. I have reported this problem before when I convert ex29 from C to Fortran. It works in Vs2008 windows but not in linux. I wonder if it has some bugs or compatibility problem.
>>>     Then likely it is an Intel compiler bug. Does this happen with the gnu gfortran compiler combined with gcc as the c compiler?
>>>
>>>    Barry
>> I have not tested. I'm also using intel fortran with VS2008 in windows 7 64bit. Do I have to re-build PETSc to test with gcc and gfortran?
>    Yes. The intel compilers will be slightly different on Linux and Windows and may have bug on linux but not on windows.
>
>    Barry

Yup, gfortran + gcc works. So do I submit a bug report to Intel?

I've heard that gfortran is slower compared to ifort. Is this true?


>
>> Thanks.
>>>>> On Jun 18, 2012, at 9:21 PM, TAY wee-beng wrote:
>>>>>
>>>>>> On 18/6/2012 9:52 PM, Matthew Knepley wrote:
>>>>>>> On Mon, Jun 18, 2012 at 7:30 PM, TAY wee-beng<zonexo at gmail.com>    wrote:
>>>>>>>
>>>>>>> On 17/6/2012 2:33 PM, Jed Brown wrote:
>>>>>>>> On Sun, Jun 17, 2012 at 7:26 AM, TAY wee-beng<zonexo at gmail.com>    wrote:
>>>>>>>>
>>>>>>>> On 16/6/2012 9:24 AM, Jed Brown wrote:
>>>>>>>>> It depends how you want to solve the problem. I usually group all dofs together. There is a 2D Stokes+Thermodynamics example in SNES ex30 (or 31?).
>>>>>>>>>
>>>>>>>> I tried to understand ex30. I have some questions since it's in C and I'm using to Fortran programming.
>>>>>>>>
>>>>>>>> This looks about right, see src/dm/examples/tutorials/ex11f90.F.
>>>>>>> I tried to build and run ex11f90 in linux. However, it's not working:
>>>>>>>
>>>>>>> I have attached the run and valgrind output:
>>>>>>>
>>>>>>> It appears to be a problem with your MPI.
>>>>>>>
>>>>>>>     Matt
>>>>>> Is there anyway to solve this problem? I'm running on a cluster using openmpi 1.5.3 and I've no admin rights.
>>>>>>> run output:
>>>>>>>
>>>>>>> [wtay at hpc12:tutorials]$ ./ex11f90
>>>>>>> [0]PETSC ERROR: ------------------------------------------------------------------------
>>>>>>> [0]PETSC ERROR: Caught signal number 11 SEGV: Segmentation Violation, probably memory access out of range
>>>>>>> [0]PETSC ERROR: Try option -start_in_debugger or -on_error_attach_debugger
>>>>>>> [0]PETSC ERROR: or see http://www.mcs.anl.gov/petsc/documentation/faq.html#valgrind[0]PETSC ERROR: or try http://valgrind.org on GNU/linux and Apple Mac OS X to find memory corruption errors
>>>>>>> [0]PETSC ERROR: likely location of problem given in stack below
>>>>>>> [0]PETSC ERROR: ---------------------  Stack Frames ------------------------------------
>>>>>>> [0]PETSC ERROR: Note: The EXACT line numbers in the stack are not available,
>>>>>>> [0]PETSC ERROR:       INSTEAD the line number of the start of the function
>>>>>>> [0]PETSC ERROR:       is given.
>>>>>>> [0]PETSC ERROR: --------------------- Error Message ------------------------------------
>>>>>>> [0]PETSC ERROR: Signal received!
>>>>>>> [0]PETSC ERROR: ------------------------------------------------------------------------
>>>>>>> [0]PETSC ERROR: Petsc Development HG revision: f3b998b41b349e16d47fe42b0e223d3462737e05  HG Date: Fri Jun 15 17:50:32 2012 -0500
>>>>>>> [0]PETSC ERROR: See docs/changes/index.html for recent updates.
>>>>>>> [0]PETSC ERROR: See docs/faq.html for hints about trouble shooting.
>>>>>>> [0]PETSC ERROR: See docs/index.html for manual pages.
>>>>>>> [0]PETSC ERROR: ------------------------------------------------------------------------
>>>>>>> [0]PETSC ERROR: ./ex11f90 on a petsc-3.3 named hpc12 by wtay Tue Jun 19 03:22:52 2012
>>>>>>> [0]PETSC ERROR: Libraries linked from /home/wtay/Lib/petsc-3.3-dev_shared_debug/lib
>>>>>>> [0]PETSC ERROR: Configure run at Sun Jun 17 16:51:29 2012
>>>>>>> [0]PETSC ERROR: Configure options --with-mpi-dir=/opt/openmpi-1.5.3/ --with-blas-lapack-dir=/opt/intelcpro-11.1.059/mkl/lib/em64t/ --with-debugging=1 --download-hypre=1 --prefix=/home/wtay/Lib/petsc-3.3-dev_shared_debug --known-mpi-shared=1 --with-shared-libraries
>>>>>>> [0]PETSC ERROR: ------------------------------------------------------------------------
>>>>>>> [0]PETSC ERROR: User provided function() line 0 in unknown directory unknown file
>>>>>>> --------------------------------------------------------------------------
>>>>>>> MPI_ABORT was invoked on rank 0 in communicator MPI_COMM_WORLD
>>>>>>> with errorcode 59.
>>>>>>>
>>>>>>> NOTE: invoking MPI_ABORT causes Open MPI to kill all MPI processes.
>>>>>>> You may or may not see output from other processes, depending on
>>>>>>> exactly when Open MPI kills them.
>>>>>>> --------------------------------------------------------------------------
>>>>>>>
>>>>>>> valgrind output:
>>>>>>>
>>>>>>> ==6027== Memcheck, a memory error detector
>>>>>>> ==6027== Copyright (C) 2002-2011, and GNU GPL'd, by Julian Seward et al.
>>>>>>> ==6027== Using Valgrind-3.7.0 and LibVEX; rerun with -h for copyright info
>>>>>>> ==6027== Command: ex11f90
>>>>>>> ==6027==
>>>>>>> ==6027== Invalid read of size 8
>>>>>>> ==6027==    at 0xA60148D: _wordcopy_fwd_dest_aligned (in /lib64/libc-2.12.so)
>>>>>>> ==6027==    by 0xA5FB11D: __GI_memmove (in /lib64/libc-2.12.so)
>>>>>>> ==6027==    by 0xA6027DB: argz_insert (in /lib64/libc-2.12.so)
>>>>>>> ==6027==    by 0x95CAF25: lt_argz_insert (ltdl.c:1679)
>>>>>>> ==6027==    by 0x95CB7D0: foreachfile_callback (ltdl.c:1718)
>>>>>>> ==6027==    by 0x95CB4F1: foreach_dirinpath (ltdl.c:710)
>>>>>>> ==6027==    by 0x95CB580: lt_dlforeachfile (ltdl.c:1865)
>>>>>>> ==6027==    by 0x95DB999: mca_base_component_find (mca_base_component_find.c:301)
>>>>>>> ==6027==    by 0x95DC4B0: mca_base_components_open (mca_base_components_open.c:128)
>>>>>>> ==6027==    by 0x95F7CE7: opal_paffinity_base_open (paffinity_base_open.c:112)
>>>>>>> ==6027==    by 0x95C39FE: opal_init (opal_init.c:307)
>>>>>>> ==6027==    by 0x95815A4: orte_init (orte_init.c:78)
>>>>>>> ==6027==  Address 0xb39d9a8 is 40 bytes inside a block of size 43 alloc'd
>>>>>>> ==6027==    at 0x4C267BA: malloc (vg_replace_malloc.c:263)
>>>>>>> ==6027==    by 0x95CA358: lt__malloc (lt__alloc.c:54)
>>>>>>> ==6027==    by 0x95CB751: foreachfile_callback (ltdl.c:1764)
>>>>>>> ==6027==    by 0x95CB4F1: foreach_dirinpath (ltdl.c:710)
>>>>>>> ==6027==    by 0x95CB580: lt_dlforeachfile (ltdl.c:1865)
>>>>>>> ==6027==    by 0x95DB999: mca_base_component_find (mca_base_component_find.c:301)
>>>>>>> ==6027==    by 0x95DC4B0: mca_base_components_open (mca_base_components_open.c:128)
>>>>>>> ==6027==    by 0x95F7CE7: opal_paffinity_base_open (paffinity_base_open.c:112)
>>>>>>> ==6027==    by 0x95C39FE: opal_init (opal_init.c:307)
>>>>>>> ==6027==    by 0x95815A4: orte_init (orte_init.c:78)
>>>>>>> ==6027==    by 0x95432AE: ompi_mpi_init (ompi_mpi_init.c:350)
>>>>>>> ==6027==    by 0x955938F: PMPI_Init (pinit.c:84)
>>>>>>> ==6027==
>>>>>>> ==6027== Syscall param writev(vector[...]) points to uninitialised byte(s)
>>>>>>> ==6027==    at 0xA65692B: writev (in /lib64/libc-2.12.so)
>>>>>>> ==6027==    by 0xC9A2C56: mca_oob_tcp_msg_send_handler (oob_tcp_msg.c:249)
>>>>>>> ==6027==    by 0xC9A417C: mca_oob_tcp_peer_send (oob_tcp_peer.c:204)
>>>>>>> ==6027==    by 0xC9A67FC: mca_oob_tcp_send_nb (oob_tcp_send.c:167)
>>>>>>> ==6027==    by 0xC3953B5: orte_rml_oob_send (rml_oob_send.c:136)
>>>>>>> ==6027==    by 0xC3955FF: orte_rml_oob_send_buffer (rml_oob_send.c:270)
>>>>>>> ==6027==    by 0xCDB1E87: modex (grpcomm_bad_module.c:573)
>>>>>>> ==6027==    by 0x95436F1: ompi_mpi_init (ompi_mpi_init.c:682)
>>>>>>> ==6027==    by 0x955938F: PMPI_Init (pinit.c:84)
>>>>>>> ==6027==    by 0x8AB4FF4: MPI_INIT (pinit_f.c:75)
>>>>>>> ==6027==    by 0x50ED97E: petscinitialize_ (zstart.c:299)
>>>>>>> ==6027==    by 0x40881C: MAIN__ (ex11f90.F:43)
>>>>>>> ==6027==  Address 0xed30cd1 is 161 bytes inside a block of size 256 alloc'd
>>>>>>> ==6027==    at 0x4C268B2: realloc (vg_replace_malloc.c:632)
>>>>>>> ==6027==    by 0x95C4F22: opal_dss_buffer_extend (dss_internal_functions.c:63)
>>>>>>> ==6027==    by 0x95C5A64: opal_dss_copy_payload (dss_load_unload.c:164)
>>>>>>> ==6027==    by 0x95A1246: orte_grpcomm_base_pack_modex_entries (grpcomm_base_modex.c:861)
>>>>>>> ==6027==    by 0xCDB1E3C: modex (grpcomm_bad_module.c:563)
>>>>>>> ==6027==    by 0x95436F1: ompi_mpi_init (ompi_mpi_init.c:682)
>>>>>>> ==6027==    by 0x955938F: PMPI_Init (pinit.c:84)
>>>>>>> ==6027==    by 0x8AB4FF4: MPI_INIT (pinit_f.c:75)
>>>>>>> ==6027==    by 0x50ED97E: petscinitialize_ (zstart.c:299)
>>>>>>> ==6027==    by 0x40881C: MAIN__ (ex11f90.F:43)
>>>>>>> ==6027==    by 0x4087AB: main (in /home/wtay/Codes/petsc-dev/src/dm/examples/tutorials/ex11f90)
>>>>>>> ==6027==
>>>>>>> ==6027== Invalid read of size 8
>>>>>>> ==6027==    at 0x50FC6D8: vecview_ (zvectorf.c:56)
>>>>>>> ==6027==    by 0x408A05: MAIN__ (ex11f90.F:56)
>>>>>>> ==6027==    by 0xEE1CC2F: ???
>>>>>>> ==6027==    by 0xEE5A0AF: ???
>>>>>>> ==6027==    by 0x6F5C9F: ??? (in /home/wtay/Codes/petsc-dev/src/dm/examples/tutorials/ex11f90)
>>>>>>> ==6027==    by 0x4962FF: ??? (in /home/wtay/Codes/petsc-dev/src/dm/examples/tutorials/ex11f90)
>>>>>>> ==6027==    by 0x6F5C9F: ??? (in /home/wtay/Codes/petsc-dev/src/dm/examples/tutorials/ex11f90)
>>>>>>> ==6027==    by 0x7FEFFFC13: ???
>>>>>>> ==6027==  Address 0xfffffffffffffeb8 is not stack'd, malloc'd or (recently) free'd
>>>>>>> ==6027==
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>> Supposed I have a field - u,v,w,p, so in order to use them, I do the following:
>>>>>>>>
>>>>>>>> type field
>>>>>>>>
>>>>>>>> real u,v,w,p
>>>>>>>>
>>>>>>>> end type field
>>>>>>>>
>>>>>>>> type(field), pointer :: field1(:,:,:)    ->    make a derived variable
>>>>>>>>
>>>>>>>> Also:
>>>>>>>>
>>>>>>>> Vec field_local,field_global
>>>>>>>>
>>>>>>>> call DMDACreate3d with dof = 4
>>>>>>>>
>>>>>>>> call DMCreateGlobalVector(da, field_local,ierr)
>>>>>>>>
>>>>>>>> call DMCreateLocalVector(da,field_global,ierr)
>>>>>>>>
>>>>>>>> call DMGetLocalVector(da, field_local,ierr)    ->       To insert values
>>>>>>>>
>>>>>>>> call DMDAVecGetArrayF90(da, field_local,field1,ierr)
>>>>>>>>
>>>>>>>> do k = zs, zs + zm - 1
>>>>>>>>
>>>>>>>> do j = ys, ys + ym -1
>>>>>>>>
>>>>>>>>      do i = xs, xs + xm - 1
>>>>>>>>
>>>>>>>>          field1(i,j,k)%u = ...     ->       evaluate u,v,w,p etc
>>>>>>>>
>>>>>>>>      end do
>>>>>>>>
>>>>>>>> end do
>>>>>>>>
>>>>>>>> call DMDAVecRestoreArrayF90(da,field_local,field1,ierr)
>>>>>>>>
>>>>>>>> call DMLocalToGlobalBegin(da,field_local,INSERT_VALUES,field_global,ierr)
>>>>>>>>
>>>>>>>> call DMLocalToGlobalEnd(da,field_local,INSERT_VALUES,field_global,ierr)
>>>>>>>>
>>>>>>>> call DMRestoreLocalVector(da,field_local,ierr)
>>>>>>>>
>>>>>>>> Is this the correct way?
>>>>>>>>
>>>>>>>> Also, supposed I now want to solve my u,v,w momentum eqns. Although they're not coupled together, I believe it's faster if I assemble them into 1 big matrix.
>>>>>>>>
>>>>>>>> So for Ax = b, x = (field(1,1,1)%u,field(1,1,1)%v,field(1,1,1)%w,field(2,1,1)%u.... )
>>>>>>>>
>>>>>>>> For b, do I duplicate a Vec similar to field_local?
>>>>>>>>
>>>>>>>> What about matrix A? Do I use the MatSetValuesStencil?
>>>>>>>>
>>>>>>>> Yes
>>>>>>>>
>>>>>>>>
>>>>>>>> Lastly, the type field contains u,v,w and p. However, I'm only solving u,v,w. Do I have to skip some values or use identity matrix to solve it?
>>>>>>>>
>>>>>>>> Why not make a field that contains only u,v,w. I don't see what you're trying to do.
>>>>>>> -- 
>>>>>>> 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


More information about the petsc-users mailing list