Re: [Libmesh-users] libmesh compilation problems on debian amd64

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Hi, many thanks, this works. Could you please create me some login to the wiki,
so that I can create a faq for this? In the meantime, this is my line:

config/configure.py --with-cc=gcc --with-cxx=g++ --with-fc=g77
--with-mpi-dir=/usr/lib/mpich/ --with-debugging=1 --with-shared

Ondrej

On Sun, Jan 01, 2006 at 03:54:17PM -0600, John Peterson wrote:
> This is definitely getting to be a FAQ.  LibMesh will try to
> get its MPI configuration from PETSc if you have petsc.  We
> need a way of disabling MPI in libmesh if it is disabled in
> PETSc...
> 
> For the moment, if you do have MPI try building petsc with MPI,
> but configuring it with the following options:
> 
> ./config/configure.py --with-cc=gcc --with-cxx=g++ --with-fc=g77 \
> --with-blas-lapack-dir=/cfdlab/Intel/MKL/mkl \
> --with-mpi-dir=/cfdlab/Mpi/1.2.7 --with-debugging=1 \
> --with-shared -RSHCOMMAND=ssh
> 
> Of course, replace the directories with your directories.  The key
> is to force PETSc to use g++/g77 instead of the mpiCC and mpif77
> wrapper compilers.  Good luck!
> 
> 
> 
> -John
> 
> 
> 
> Ondrej Certik writes:
>  > Hello, I wasn't able to compile libmesh (tried cvs, 0.5.0) with petsc (tried
>  > 2.2.1, 2.3.0) on debian sid on amd64. It compiles ok without petsc. 
>  > I tried a lot of combinations (different versions, with mpi, without...) but
>  > all give more or less the same kind of errors, so I guess we can concentrate on
>  > this setup:
>  > 
>  > libmesh cvs, petsc 2.3.0, gcc-4.0:
>  > 
>  > cvs -d:pserver:ano...@cv...:/cvsroot/libmesh co libmesh
>  > cd libmesh
>  > cd contrib
>  > wget ftp://ftp.mcs.anl.gov/pub/petsc/petsc-lite.tar.gz
>  > tar xzf petsc-lite.tar.gz
>  > cd petsc-2.3.0
>  > export PETSC_ARCH=linux
>  > export PETSC_DIR=`pwd`
>  > config/configure.py --with-mpi=0
>  > make
>  > cd ..
>  > ./configure 
>  > make
>  > 
>  > am I doing something wrong here? It stops here:
>  > 
>  > src/mesh/mesh_communication.C: In member function 'void
>  > MeshCommunication::broadcast_bcs(MeshBase&, BoundaryInfo&) const':
>  > src/mesh/mesh_communication.C:418: error: 'MPI_UNSIGNED_SHORT' was not declared
>  > in this scope
>  > 
>  > The problem is, that the file include/base/libmesh_config.h contains
>  > 
>  > #define HAVE_MPI 1
>  > 
>  > (it doesn't help passing --disable-mpi to ./configure). Basically, we have 2
>  > options - either I comment the line HAVE_MPI in libmesh_config.h, or I define
>  > MPI_UNSIGNED_SHORT somewhere (the problem is in mesh_communication.C and in
>  > some other file I forgot which). Let's put this line for example to
>  > include/mesh/mesh_base.h:
>  > 
>  > #define MPI_UNSIGNED_SHORT   sizeof(unsigned short)
>  > 
>  > Then the make compiles libmesh without errors (well, I have to disable
>  > parmetis, which complains like this:
>  > 
>  > --- Building Parmetis ------------------------
>  > make[2]: Entering directory
>  > `/home/ondra/libmesh/cvslibmesh/libmesh/contrib/parmetis/Lib'
>  > Compiling C (in optimized mode) akwayfm.c...
>  > In file included from ././stdheaders.h:26,
>  >                  from ./parmetislib.h:19,
>  >                  from akwayfm.c:14:
>  > /home/ondra/libmesh/cvslibmesh/libmesh/contrib/petsc-2.3.0/include/mpiuni/mpi.h:
>  > 61:23: error: petscconf.h: No such file or directory
>  > make[2]: *** [akwayfm.x86_64-unknown-linux-gnu.opt.o] Error 1
>  >                                   
>  > ), but
>  > 
>  > make run_examples 
>  > 
>  > fails in ex0:
>  > 
>  > ***************************************************************
>  > * Running Example  ./ex0
>  > ***************************************************************
>  > 
>  > --------------------------------------------------------------------------
>  > Petsc Release Version 2.3.0, Patch 43, April, 26, 2005
>  > See docs/changes/index.html for recent updates.
>  > See docs/faq.html for hints about trouble shooting.
>  > See docs/index.html for manual pages.
>  > -----------------------------------------------------------------------
>  > Libraries linked from
>  > /home/ondra/libmesh/cvslibmesh/libmesh/contrib/petsc-2.3.0/lib/linux
>  > Configure run at Sun Jan  1 14:23:17 2006
>  > Configure options --with-mpi=0 --with-shared=0
>  > -----------------------------------------------------------------------
>  > [-1]PETSC ERROR: PetscInitialize() line 399 in src/sys/src/objects/pinit.c
>  > [-1]PETSC ERROR: No support for this operation for this object type!
>  > [-1]PETSC ERROR: You cannot set PETSC_COMM_WORLD if you have not initialized
>  > MPI first!
>  > [-1]PETSC ERROR: User provided function() line 166 in
>  > unknowndirectory/src/base/libmesh.C
>  > 
>  > So I guess the problem was, that petsc was compiled without MPI, but libmesh
>  > with MPI, so I uncomment the HAVE_MPI in libmesh_config, then I got:
>  > 
>  > /home/ondra/libmesh/cvslibmesh/libmesh/include/numerics/petsc_vector.h: In
>  > member function 'void PetscVector<T>::init(unsigned int, unsigned int, bool)':
>  > /home/ondra/libmesh/cvslibmesh/libmesh/include/numerics/petsc_vector.h:523:
>  > error: 'COMM_WORLD' is not a member of 'libMesh'
>  > /home/ondra/libmesh/cvslibmesh/libmesh/include/numerics/petsc_vector.h:525:
>  > error: 'COMM_WORLD' is not a member of 'libMesh'
>  > 
>  > and a lots of errors like that. The problem is obvious - the COMM_WORLD in the
>  > namespace libMesh is defined only if we have HAVE_MPI defined. So I guess I
>  > cannot compile the libmesh without MPI at all (if I tried to fake it somehow,
>  > there were other errors with MPI - it seems to me, that I need MPI enabled in
>  > libmesh in order to use petsc).
>  > 
>  > So I didn't succeed without MPI, let's try with it. If I compile petsc with mpi
>  > and try libmesh, I get an error right at the beginning, that it cannot find
>  > mpi.h - although I passed an option to configure --with-mpi=/usr/lib/mpich
>  > (which contains the mpi.h). I am a beginner with MPI, and I don't need it.  Is
>  > it possible to compile libmesh without MPI? If not, could please someone give
>  > me a hint how to compile it with MPI? I would be glad to compile libmesh anyhow
>  > with petsc (and later with slepc). I think I am doing something wrong, maybe
>  > the petsc should be compiled differently?
>  > 
>  > BTW 1:
>  > if you want to use slepc 2.3.0, you need to make some changes in Make.common
>  > around the line 291:
>  > comment the line 291, change another line to: 
>  > SLEPC_LIB = $(SLEPC_DIR)/lib/$(PETSC_ARCH)
>  > 
>  > BTW 2: do only me have these problems with cvs?
>  > 
>  > ondra@dakol:~/libmesh/cvslibmesh$ cvs -d:pserver:ano...@cv...:/cvsroot/libmesh co libmesh
>  > cvs checkout: CVS password file /home/ondra/.cvspass does not exist - creating
>  > a new file 
>  > cvs [checkout aborted]: end of file from server (consult above messages if any)
>  > 
>  > This happens maybe 30 times in a row, the 31st try starts downloading...  
>  > 
>  > Ondrej
>  > 
>  > 
>  > -------------------------------------------------------
>  > This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
>  > for problems?  Stop!  Download the new AJAX search engine that makes
>  > searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
>  > http://ads.osdn.com/?ad_id=7637&alloc_id=16865&op=click
>  > _______________________________________________
>  > Libmesh-users mailing list
>  > Lib...@li...
>  > https://lists.sourceforge.net/lists/listinfo/libmesh-users
> 
> -- 
> John W. Peterson
> The University of Texas at Austin
> 
> 
> Web:    http://www.cfdlab.ae.utexas.edu
> Phone:  512-619-6534 (H)  |  512-471-4069 (W)
> email:  pet...@cf...