Skip to content

Compiling and installing

Jump to bottom
Thierry Fredrich edited this page Jun 25, 2019 · 21 revisions

Building

Requirements

Successfully built and run on

  • 4.1.11-gentoo-ARCH
  • 4.1.7-hardened-r1-ARCH
  • SMP Debian 3.2.68-1+deb7u5 x86_64 GNU/Linux
  • Ubuntu 16.04 4.4.0-45-generic #66-Ubuntu
  • Ubuntu 14.04 4.2.0-27-generic #32

Build process

CMake is our build system. You most likely need to configure the build according to your system specifics. To do so, we recommend you to create a shell script which contains the call to cmake while storing all your personal options. We have several examples in the install_helpers directory.

An example usage of our install_helpers on Ubuntu 16.04 works as follows:

export CC=mpicc #set c compiler to mpi
export CXX=mpicxx #sets cxx compiler to mpi
$mkdir build  # create your build directory
$cd build
$./ /path/to/source/directory/tumorcode/install_helpers/do-configure-tumorcode-ubuntu16.sh /path/to/source/directory/tumorcode
$make
$make install

Basic CMake usage is as follows

$mkdir build  # create your build directory
$cd build
$cmake -DCMAKE_INSTALL_PREFIX=/home/thierry/tc_install path_to_tumorcode
$make
$make install

The install command is required since it copies generated python libraries to their proper place within the py subdirectory. Note that you can set CMAKE_INSTALL_PREFIX to your source directory, which is useful for development. Note that /home/thierry/tc_install is only an example here. Using the curses based interface ccmake might also be an option to configure your builds.

Details

Here is a list of debian packages you will need: sudo apt-get install cmake gfortran libblas-dev liblapack-dev python2.7-dev libboost-all-dev libboost-python-dev python-numpy python-matplotlib libpng-dev libtbb-dev libeigen3-dev python-h5py python-scipy povray mpi-default-dev libmumps-dev libsuperlu-dev libhdf5-dev libhdf5-openmpi-dev libptscotch-dev binutils-dev libiberty-dev libtrilinos-* cimg-dev libcgal-dev libcgal-qt5-dev libann-dev

Our HDF5-cpp-wrapper is a lightweight c++ library around HDF5. Get it by git clone https://github.com/DaWelter/HDF5-cpp-wrapper.git As it is only a single-header you can simply put a link to it in a directory in your include paths.

NumpyCPP is our C++ wrapper around the numpy C API. Get it by git clone https://github.com/DaWelter/numpycpp.git. Compile and install by

$cd numpycpp
$mkdir build
$cmake -DCMAKE_INSTALL_PREFIX=/home/thierry/tc_install ../
$make
$make install

Installing

In order to use the command system wide you have to export the binary path to your system. With the bash system you can do this by adding the following line to your .bashrc file in the home folder.

export PATH=$PATH:/home/thierry/tc_install/bin

Debug Mode

We followed the python standards (https://docs.python.org/2.7/c-api/intro.html#debugging-builds), we designed the python interface of tumorcode in such a way that debugging builds are invoked with the -d flag of python. We give an example with holds true for all other cases. To run the vesselgenerator in debug mode execute:

python2 -d /home/thierry/tc_install/py/krebsjobs/submitVesselgeneration.py -h

Issues

Linking problems with HDF5 and h5py

17.11.17 we intended to remove this odd dependency for the current master branch. For reference we leave this information here.

IMPORTANT: We fetch object references pertaining to the HDF5 library linked with h5py and use functions on them pertaining to the HDF5 lib linked to our code, assuming both libraries are the same. THIS MIGHT, HOWEVER, NOT BE THE CASE.

If so, on Ubuntu 16 for instance, you have to compile h5py by yourself. Here is how you can achieve this:

Install the requirements sudo apt-get install cython python-mpi4py python-pkgconfig. Then get the h5py source, e.g. from GitHub git clone https://github.com/h5py/h5py. Then build and install by

$cd h5py
$python2 setup.py configure --hdf5=/usr/include/hdf5/openmpi --mpi
$python2 setup.py build
$sudo python2 setup.py install
# or if you want to install it in a custom directory
$python2 setup.py install --prefix=target_dir

By --hdf5=... and --mpi we request to link against the hdf5 library that is installed by respective debian packages, which happen to be compiled with MPI Support.

MPI

This software performs all computations locally, as opposed to distributed HPC. Therefore MPI is NOT required. However, it should do no harm if trilinos and other packages are linked to MPI.

Ubuntu 14.04

Since trilinos was added to ubuntu in 16.04 you have add it manually in previous versions. We recommend: https://launchpad.net/~nschloe/+archive/ubuntu/trilinos-nightly/

Packages

Note that the labeling and versions of the packages vary slightly between Ubuntu 14.04 and 16.04. Here is a list of debian packages you will need: sudo apt-get install cmake gfortran libblas-dev liblapack-dev python2.7-dev libboost-all-dev libboost-python-dev python-numpy python-matplotlib libboost-program-options-dev libpng12-dev libtbb-dev libeigen3-dev python-h5py python-scipy povray mpi-default-dev libmumps-dev libsuperlu3-dev libhdf5-openmpi-dev libptscotch-dev binutils-dev libiberty-dev libtrilinos-* cimg-dev

BUG

Unfortunately we are experiencing this bug, (https://github.com/trilinos/Trilinos/issues/1225) when configuring tumorcode at the moment.

Compiling Trilinos without MPI

We usually completely eliminate MPI from your build to avoid any potential for trouble. Therefore we compile the Trilinos suite by ourselves. To assist people in doing the same, we provide configure scripts located in install_helpers.

For scientific information see http://www.uni-saarland.de/fak7/rieger/homepage/research/tumor/tumor.html

Clone this wiki locally