Rheolef  7.1
an efficient C++ finite element environment
Configuration

The simplest install procedure is as the root user: 

./configure
make
sudo make install

The configure program is a script that runs tests of your system and the availability of some required and optional libraries for Rheolef. Upon termination, it prints a summary and generates two files: config/config.mk and config/config.h. The make program then runs the conpilation and the make install stage copies the produced files. By default, the main destination directory is /usr/local. Commands are installed in /usr/local/bin, libraries in /usr/local/lib, headers in /usr/local/include and the documentation in /usr/local/share.

Alternative install directory

Since the default install prefix is /usr/local, it requires to be logged as root in order to install. If you are not the root user, or prefer an alternative install directory, e.g. $HOME/sys, enter: 

./configure --prefix="$HOME/sys"

In that case, you have to add the following lines at the end of your .profile or .bash_profile file: 

export PATH="$HOME/sys/bin:$PATH"
export LD_LIBRARY_PATH="$HOME/sys/lib:$LD_LIBRARY_PATH"
export MANPATH="$HOME/sys/man:$MANPATH"

Assuming you are using the sh or the bash unix interpreter. Conversely, if you are using csh or tcsh, add at the bottom of your .cshrc file: 

set path = ("$HOME/sys/bin" $path)
setenv LD_LIBRARY_PATH "$HOME/sys/lib:$LD_LIBRARY_PATH"
setenv MANPATH "$HOME/sys/man:$MANPATH"

If you are unsure about unix interpreter, get the SHELL value: 

echo $SHELL

Then, you have to source the modified file, e.g.: 

source ~/.bashrc

HPUX users should replace LD_LIBRARY_PATH by SHLIB_PATH. If you are unsure, get the shared library path variable: 

rheolef-config --shlibpath-var

Check also your installation by compiling a sample program with the installed library: 

make installcheck

Since it is a common mistake, you can later, at each time, test your run-time environment sanity with: 

rheolef-config --check

Required libraries

The following required libraries are available as precompiled packages under Debian GNU/Linux and Ubuntu systems.

The boost library. It provides useful extensions of the c++ standard template library. See http://www.boost.org .

The eigen library. It provides efficient c++ linear algebra subroutines. See http://eigen.tuxfamily.org .

Recommended libraries

The following libraries are optional but highly recommended.

The umfpack library. It provides a multifrontal sparse direct solver. Rheolef solvers run faster when using this library. See http://faculty.cse.tamu.edu/davis/suitesparse.html .

The cgal library. It provide an easy access to efficient and reliable geometric algorithms. With it, Rheolef implements the characteristic method and interpolation from one mesh to another. This library is optional, but allows, when available, extended Rheolef features. See http://www.cgal.org .

Parallel computations

With the optional following libraries, Rheolef is able to run in parallel and and with distributed memory. By default, the Rheolef configure script tries to autodetect these libraries and be able to run in parallel. When these libraries are not found, the distributed features are turned off. If your libraries are installed in a non-standard location, you should indicate it by a suitable configure option. See below for these options.

A mpi library. Rheolef bases on mpi, the message passing interface standard. A popular implementation is the openmpi library. When no mpi library is found, the distributed feature is turned off.

The mumps library. It is a massively distributed sparse direct solver. The use of implicit numerical schemes leads to the resolution of large sparse linear systems. See http://graal.ens-lyon.fr/MUMPS . When mumps is not found, the distributed feature is turned off.

Either the scotch or the metis library. It provides the distributed mesh partitioning. For scotch see http://www.labri.fr/perso/pelegrin/scotch and for metis see http://www.cs.umn.edu/~metis . When neither scotch nor metis are founded, the distributed feature is turned off.

The trilinos/ifpack library. It provides a distributed preconditioner used for iterative solvers. See http://trilinos.sandia.gov . When trilinos is not available, a less efficient builtin IC0 preconditioner is used. The trilinos support is actually broken, it will be available in a forthcoming version.

Run-time tools

Using Rheolef suggests the use of the following optional tools for pre- and post-processing purpose: 

gmsh
gnuplot
paraview

None of these tools are required when building and installing Rheolef.

Tools for developpers

These tools are required if you build from the development version, from the git repository:

  • Re-generating some source files, as initiated with the bootstrap command, requires the autotools suite: autoconf, automake and libtool.
  • It requires also flex and bison for building file format parsers.

Portability issues

Rheolef is based on standard C++ and the boost library. Therefore you need a compiler that supports at least the 2011 C++ standards.

Current distributed version of Rheolef has been heavily tested with GNU C++, CLANG C++ and INTEL C++ on debian, ubuntu, mint, suse and redhat GNU/Linux systems for various 32 and 64 bits architectures, with massively distributed memories and processes. Successful installation from sources on MAC-OS-X systems has been also repported recently.

  • The GNU C++ compiler g++ is a free software available at http://www.gnu.org .
  • The CLANG C++ compiler clang++ is also a free software available at http://clang.llvm.org .
  • The Intel C++ compiler is a privative software.

So, the Rheolef library is expected to compile without problems with one of these compilers. If you are running a another compiler and operating system combination, please run the non-regression test suite: 

make check

Please, send mails and bug reports to the mailing list rheol.nosp@m.ef@g.nosp@m.renet.nosp@m..fr .

Older Rheolef 5.x versions was tested also on the KAI C++ compiler (KCC-3.2b) and the CRAY C++ compiler (CC-3.0.2.0) on the following operating systems:

  • mac os x (i386-apple-darwin9.8.0), using GNU C++
  • sun solaris 2.8, ultrasparc, using GNU C++
  • compacq OSF1 alpha (V5.1, V4.0), using Compacq C++
  • aix 4.1 ibm sp2, multiprocessor based on rs6000
  • cray unicos t3e, multiprocessor based on 64 bits dec alpha
  • cray unicos c94
  • hpux 9.05 and 10.20, hppa
  • sgi irix 6.3, mips
  • sun solaris 2.5.1, sparc

where

  • The KAI C++ is available at http://www.kai.com .
  • The CRAY C++ is available on CRAY platforms.

Nevertheless, the current version has no more been tested on these old compilers and systems.

Configure options

If the required libraries are located in a non-standard directory, please use the configure options: 

./configure --help

and also consider the CXX, CXXFLAGS and LIBS variables, as in the following more complex configure command: 

CXX=clang++ CXXFLAGS=-I/misc/include LIBS="-L/misc/lib" ./configure

Finaly, if troubles still persist after the configure step, you can also edit directly the config/config.mk and config/config.h files. In these cases please also post to the mailling list rheol.nosp@m.ef@g.nosp@m.renet.nosp@m..fr for any portability problem.

Here are all the supported configure command line options.

-h, --help

Display a summary of command line options and exit

--help=short

Display only options specific to Rheolef and exit

--with-float=option

Turns on/off the extended arithmetic feature. The option could be either float128 (quadruple precision, 32 digits), or the default double (double precision, 16 digits).

--enable-permissive

The compiler do not turns all warnings into errors and becomes permissive. This feature is for code developers and is off by default.

--enable-debian-packaging

Install files in the debian way by respecting the Debian Policy Manual. This is not the default behavior.

--enable-documentation

By default, both the reference manual and the users guide documentations are generated when required tools are available. The required tool for the reference manual is doxygen. The required tools for the users guide are pdflatex, bibtex, makeindex, gs, fig2dev, gnuplot, inkscape. This option allows one to turn off the generation of these documentation.

--with-bigfloat=digits10

Turn on Bruno Haible's cln arbitrary precision float library with digits10 precision, See http://clisp.cons.org/~haible/packages-cln.html . Default is off. The digits10 argument value is optional and default value is digits10=60. This feature is still under development.

--with-cln=clndir

Set the home directory for the cln library used by the --with-bigfloat previous option. Default clndir is /usr/local/math.

--enable-maintainer-mode

This option is off by default. When false, it turns off by default the ginac feature (see below). If you are not a Rheolef library code developper, you could safely turn it off. This option is provided for compatibility with the Debian packaging system debhelper.

--enable-ginac

This option is off by default. Ginac is computer algebra system in C++ that is used to automatically generate the C++ code for some low-order FEM basis (P1, bubble, etc). If you are not a Rheolef library code developper, you could safely turn it off.

--with-boost-incdir=incdir_boost

Specifies the include directory for the boost library. This library is required by rheolef.

--with-boost-libdir=libdir_boost

Specifies the corresponding library directory.

--with-eigen-incdir=incdir_eigen

Specifies the include directory for the eigen library. This library is required by rheolef.

--with-eigen-libdir=libdir_eigen

Specifies the corresponding library directory.

--enable-debug

Produce a c++ -g like code and turn on some runtime checks.

--enable-dmalloc

With debugging, also uses dynamic allocation runtime checking with dmalloc when available.

--enable-mpi

Turns on/off the distributed computation features. Default is on when distributed is on. This feature is currently in development. When on, configure try to detect either the scotch or the parmetis libraries.

--with-scotch-incdir=incdir_scotch

Specifies the include directory for the scocth distributed mesh partitioner. Default is to auto-detect when mpi is available.

--with-scotch-libdir=libdir_scotch

Specifies the corresponding library directory. Check in libdir_scotch for libptscotchparmetis.so, libptscotch.so and libptscotcherrexit.so or corresponding .a libraries.

--with-scotch-libs=libs_scotch

Specifies some alternating library compile flag e.g. -lscotch.

--with-parmetis-incdir=incdir_parmetis

Specifies the include directory for the parmetis distributed mesh partitioner. Default is to autodetect when mpi is available and scotch is unavailable.

--with-parmetis-libdir=libdir_parmetis

Specifies the corresponding library directory. Check in libdir_parmetis for libparmetis.so, libparmetis.a and also libmetis.a libmetis.so.

--with-blas-libdidr=rheo_libdir_blas

Specifies the include directory for the blas linear algebra library. Default is to auto-detect.

--with-blas-libs=rheo_libs_blas

Specifies the corresponding library directory.

--with-scalapack-libdir=rheo_libdir_scalapack

Specifies the library directory for the scalapack library. Default is to auto-detect when mpi is available.

--with-scalapack-libs=rheo_libs_scalapack

Specifies some alternating library compile flag e.g. -lscalapack.

--with-mumps-incdir=incdir_mumps

Specifies the include directory for the mumps linear algebra library. Default is to auto-detect.

--with-mumps-libdir=libdir_mumps

Specifies the corresponding library directory.

--with-mumps-libs=libs_mumps

Specifies some alternating library compile flag e.g. -lmumps.

--with-trilinos=libdir_trilinos

Specifies the library directory for the trilinos/ifpack preconditioner library. Check in libdir_trilinos for libtrilinos_ifpack.so, or the corresponding .a libraries. The trilinos support is actually broken.

--with-cholmod-incdir=incdir_cholmod

Specifies the include directory for the cholmod direct solver library for sparse symmetric definite positive matrix. Default is to auto-detect.

--with-cholmod-libdir=libdir_cholmod

Specifies the corresponding library directory. Default incdir_cholmod is libdir_cholmod\ . Check in libdir_cholmod for libcholmod.so or libcholmod.a and for header incdir_cholmod\ /cholmod.h or incdir_cholmod\ /cholmod/cholmod.h.

--with-cholmod-libs=libs_cholmod

Specifies some alternating library compile flag e.g. -lcholmod.

--with-umfpack-incdir=incdir_umfpack

Specifies the include directory for the umfpack direct solver library for sparse unsymmetric matrix. Default is to auto-detect.

--with-umfpack-libdir=libdir_umfpack

Specifies the corresponding library directory. Default incdir_umfpack is libdir_umfpack\ . Check in libdir_umfpack for libumfpack.so or libumfpack.a and for header incdir_umfpack\ /umfpack.h or incdir_umfpack\ /umfpack/umfpack.h.

--with-umfpack-libs=libs_umfpack

Specifies some alternating library compile flag e.g. -lumfpack.

--with-cgal-incdir=incdir_cgal

Specifies the include directory for the cgal discrete geometry library. Default is to auto-detect.

--with-cgal-libdir=libdir_cgal

Check in libdir_cgal for libcgal.so, or the corresponding .a libraries.

--with-cgal-libs=libs_cgal

Specifies some alternating library compile flag e.g.

-lCGAL. --enable-optim

Turns compile-time optimization to maximum available. Include architecture-dependent compilation (e.g. on intel, executable are less portable across intel variants). Default is on.