Bug#712526: freefem++: FreeFem++-mpi not working

Giuseppe Pitton giuseppe.pitton at gmail.com
Tue Jun 18 10:57:14 UTC 2013 

Dear Dimitris,
here are some thoughts on how to solve the issue (i tried to make HYPRE solver work).
Installing from source FreeFem++-3.23 I noticed that in the folder ff++/downloads/hypre nothing was built. So I compiled manually hypre-2.9.0b from source and I changed the paths in the file ff++/src/solver/makefile-sparsesolver.inc and I also commented the non HYPRE lines in ff++/src/solver/makefile. Please note that I had to modify several other lines in ff++/src/solver/makefile-sparsesolver.inc, in particular:
on line 56 after MPI_INCLUDE = there is a /I/ that should instead be a -I
on line 77 suffix should be changed to so
several paths should be adapted to the system in use (for instance lines 87, 121, 126, 130, 135 and of course lines 185 and following)
lines 99 and 100 should be commented
then moving in directory ff++/src/solver and running make I get a hypre_FreeFem.so that I moved to FreeFem++ library directory (/usr/local/lib/ff++/3.23/lib/ on my system). Now HYPRE is running but not working, see this test case for instance:

$ ff-mpirun -np 2 chaleur3d-hypre.edp 
initparallele rank 0 on 2
-- FreeFem++ v  3.230000 (date Jeu  6 jui 2013 20:58:44 CEST)
 Load: lg_fem lg_mesh lg_mesh3 eigenvalue parallelempi 
(large output)
  -- Square mesh : nb vertices  =441 ,  nb triangles = 800 ,  nb boundary edges 80
  -- Square mesh : nb vertices  =441 ,  nb triangles = 800 ,  nb boundary edges 80
  -- Build Nodes/DF on mesh :   n.v. 9261, n. elmt. 48000, n b. elmt. 4800
     nb of Nodes 68921    nb of DoF   68921  DFon=1100
  -- FESpace: Nb of Nodes 68921 Nb of DoF 68921
  -- Build Nodes/DF on mesh :   n.v. 9261, n. elmt. 48000, n b. elmt. 4800
     nb of Nodes 68921    nb of DoF   68921  DFon=1100
  -- FESpace: Nb of Nodes 68921 Nb of DoF 68921
###########DEFAULT PARAMETERS WILL BE SET#######
SOLVER: AMG-GMRES
tgv1e+10
time 0.01
tgv1e+10
time 0.01
 *** Process received signal ***
 Signal: Segmentation fault: 11 (11)
 Signal code: Address not mapped (1)
 Failing at address: 0x7f88d259d880
 *** Process received signal ***
 Signal: Segmentation fault: 11 (11)
 Signal code: Address not mapped (1)
 Failing at address: 0x7fe9f2d9d5c0
 *** End of error message ***
 *** End of error message ***
--------------------------------------------------------------------------
mpirun noticed that process rank 1 with PID 550 on node (...) exited on signal 11 (Segmentation fault: 11).
--------------------------------------------------------------------------



Hope this helps.
Best regards.
Giuseppe


Il giorno 17/giu/2013, alle ore 12:18, Dimitrios Eftaxiopoulos ha scritto:

> Hello Giuseppe
> Thank you for raising the mpi related issues. 
> The failing to load msh3 seems to me to be a more general issue not related to 
> the parallel solver in particular. 
> For the rest of the problems I do not have an immediate answer. 
> I plan to try to upload freefem++-3.23 to unstable shortly so I will have a 
> look on these issues.
> 
> Best regards
> Dimitris 
> 
> On Sunday 16 of June 2013 21:47:49 you wrote:
>> Package: freefem++
>> Version: 3.19.1-1
>> Severity: normal
>> 
>> Dear Maintainer,
>> I installed freefem++ on the default Wheezy (7.1) with apt-get, I also
>> installed freefem++-dev. I experienced the same issue on Wheezy (7.0).
>> FreeFem++ is working properly, but the parallel version FreeFem++-mpi does
>> not. I tried running the built-in examples of the examples++-mpi folder and
>> they are not working. The mentioned example files can be found at the
>> following URL: http://www.freefem.org/ff%2B%2B/ff%2B%2B/examples%2B%2B-mpi/
>> In particular, I tried running some test cases with both the following
>> commands:
>> FreeFem++-mpi chaleur3D-hips.edp
>> ff-mpirun -np 2 chaleur3D-mumps.edp
>> I obtain two kind of errors: msh3 libraries are missing or parallel solver
>> libraries are missing, for instance:
>> 
>> $ FreeFem++-mpi chaleur3D-pastix.edp
>> initparallele rank 0 on 1
>> -- FreeFem++ v  3.190001 (date Mer  9 mai 2012 21:50:21 CEST)
>> Load: lg_fem lg_mesh lg_mesh3 eigenvalue parallelempi
>>    1 :  // other
>>    2 : load "msh3"
>> load error : msh3
>>         fail :
>> list  prefix: './' '/usr/lib/x86_64-linux-gnu/freefem++/' list  suffix : ''
>> , '.so'
>> 
>> Error line number 2, in file chaleur3D-pastix.edp, before  token msh3
>> Error load
>>  current line = 2 mpirank 0 / 1
>> Compile error : Error load
>>        line number :2, msh3
>> error Compile error : Error load
>>        line number :2, msh3
>> code = 1 mpirank: 0
>> FreeFem++-mpi finalize correctly .
>> 
>> $ FreeFem++-mpi chaleur3D-superludist.edp
>> initparallele rank 0 on 1
>> -- FreeFem++ v  3.190001 (date Mer  9 mai 2012 21:50:21 CEST)
>> Load: lg_fem lg_mesh lg_mesh3 eigenvalue parallelempi
>>    1 :  // other
>>    2 :  // NBPROC 2
>>    3 : // ff-mpirun -np 4 chaleur3D-superludist.edp -glut ffglut  -n 20 -op
>> 1 -dt 0.01 -niter 10
>>    4 :
>>    5 : load "real_SuperLU_DIST_FreeFem"
>> load error : real_SuperLU_DIST_FreeFem
>>         fail :
>> list  prefix: './' '/usr/lib/x86_64-linux-gnu/freefem++/' list  suffix : ''
>> , '.so'
>> 
>> Error line number 5, in file chaleur3D-superludist.edp, before  token
>> real_SuperLU_DIST_FreeFem
>> Error load
>>  current line = 5 mpirank 0 / 1
>> Compile error : Error load
>>        line number :5, real_SuperLU_DIST_FreeFem
>> error Compile error : Error load
>>        line number :5, real_SuperLU_DIST_FreeFem
>> code = 1 mpirank: 0
>> FreeFem++-mpi finalize correctly .
>> 
>> I partially solved these issues compiling from source FreeFem++ v.3.23 with
>> the configure command:
>> ../configure '--enable-download'
>> then make and make install as usual.
>> This however does not solve completely the problem since the make check
>> gives 80 out of 84 test passed.
>> 
>> I have gcc 4.7.2 and openmpi 1.4.5.
>> I made HIPS, MUMPS and Super LU parallel solvers work, while HYPRE, pARMS
>> and pastix still do not. Here is one example of a working parallel
>> simulation with HIPS:
>> 
>> $ ff-mpirun -np 2 chaleur3D-hips.edp
>> '/opt/openmpi/1.6.4/bin/mpirun' -np 2 /usr/local/bin/FreeFem++-mpi
>> chaleur3D- hips.edp
>> initparallele rank 0 on 2
>> -- FreeFem++ v  3.230000 (date Sun Jun 16 19:29:19 CEST 2013)
>> Load: lg_fem lg_mesh lg_mesh3 eigenvalue parallelempi
>> (large output...)
>>   ~Hips_Solver S:0
>> times: compile 0.01s, execution 17.71s,  mpirank:1
>> times: compile 0.01s, execution 18.46s,  mpirank:0
>> ######## We forget of deleting   3 Nb pointer,   0Bytes  ,  mpirank 0
>> CodeAlloc : nb ptr  3401,  size :382192 mpirank: 0
>> Bien: On a fini Normalement
>> ######## We forget of deleting   3 Nb pointer,   0Bytes  ,  mpirank 1
>> CodeAlloc : nb ptr  3401,  size :382192 mpirank: 1
>> FreeFem++-mpi finalize correctly .
>> FreeFem++-mpi finalize correctly .
>> 
>> and an example of a failing simulation with HYPRE:
>> 
>> $ ff-mpirun -np 2 chaleur3D-hypre.edp
>> '/opt/openmpi/1.6.4/bin/mpirun' -np 2 /usr/local/bin/FreeFem++-mpi
>> chaleur3D- hypre.edp
>> initparallele rank 0 on 2
>> -- FreeFem++ v  3.230000 (date Sun Jun 16 19:29:19 CEST 2013)
>> Load: lg_fem lg_mesh lg_mesh3 eigenvalue parallelempi
>>    1 :
>>  current line = 5 mpirank 1 / 2
>>    2 :  // other
>>    3 : load "msh3" (load: dlopen ../examples++-load/msh3.so 0x1416e60) 
>> load: msh3
>> 
>>    4 : load "medit" (load: dlopen ../examples++-load/medit.so 0x14187f0)
>>    5 : load "hypre_FreeFem"
>> load error : hypre_FreeFem
>>         fail :
>> list  prefix: '../examples++-load/' '' './' list  suffix : '' , '.so'
>> 
>> Error line number 5, in file chaleur3D-hypre.edp, before  token
>> hypre_FreeFem Error load
>>  current line = 5 mpirank 0 / 2
>> Compile error : Error load
>>        line number :5, hypre_FreeFem
>> error Compile error : Error load
>>        line number :5, hypre_FreeFem
>> code = 1 mpirank: 0
>> FreeFem++-mpi finalize correctly .
>> --------------------------------------------------------------------------
>> mpirun noticed that the job aborted, but has no info as to the process
>> that caused that situation.
>> --------------------------------------------------------------------------
>> 
>> The other non-working solvers give the same error.
>> 
>> 
>> 
>> 
>> -- System Information:
>> Debian Release: 7.1
>>  APT prefers stable
>>  APT policy: (500, 'stable')
>> Architecture: amd64 (x86_64)
>> Foreign Architectures: i386
>> 
>> Kernel: Linux 3.2.0-4-amd64 (SMP w/2 CPU cores)
>> Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8)
>> Shell: /bin/sh linked to /bin/dash
>> 
>> Versions of packages freefem++ depends on:
>> ii  freeglut3                  2.6.0-4
>> ii  libamd2.2.0                1:3.4.0-3
>> ii  libarpack2                 3.1.1-2.1
>> ii  libblas3 [libblas3gf]      1.2.20110419-5
>> ii  libblas3gf                 1.2.20110419-5
>> ii  libc6                      2.13-38
>> ii  libgcc1                    1:4.7.2-5
>> ii  libgfortran3               4.7.2-5
>> ii  libgl1-mesa-glx [libgl1]   8.0.5-4+deb7u1
>> ii  libglu1-mesa [libglu1]     8.0.5-4+deb7u1
>> ii  liblapack3 [liblapack3gf]  3.4.1+dfsg-1+deb70u1
>> ii  libopenmpi1.3              1.4.5-1
>> ii  libstdc++6                 4.7.2-5
>> ii  libumfpack5.4.0            1:3.4.0-3
>> ii  libx11-6                   2:1.5.0-1+deb7u1
>> ii  libxext6                   2:1.3.1-2+deb7u1
>> ii  libxpm4                    1:3.5.10-1
>> ii  libxxf86vm1                1:1.1.2-1+deb7u1
>> 
>> freefem++ recommends no packages.
>> 
>> freefem++ suggests no packages.

More information about the debian-science-maintainers mailing list