opm-autodiff contains a small library for automatic differentiation built on the Eigen linear algebra package. It also contains some experimental solver and simulator prototypes demonstrating how it can be used. The most important parts are:
- AutoDiffBlock.hpp (class for AD on vectorized data with sparse jacobians)
- AutoDiffHelpers.hpp (various utilities to simplify writing solvers)
- sim_fibo_ad.cpp (prototype fully implicit black-oil simulator)
The library is distributed under the GNU General Public License, version 3 or later (GPLv3+).
The opm-autodiff module is designed to run on Linux platforms. It is also regularly run on Mac OS X. No efforts have been made to ensure that the code will compile and run on windows platforms.
opm-autodiff requires opm-core, and all its requirements (see opm-core/README). In addition, opm-autodiff requires Eigen, version 3.1 (has not been tested with later versions).
For a read-only download: git clone git://github.com/OPM/opm-autodiff.git
If you want to contribute, fork OPM/opm-autodiff on github.
See build instructions in opm-core/README, or at http://opm-project.org/download.php.
Efforts have been made to document the code with Doxygen. In order to build the documentation, enter the command
make doc
in the topmost directory. The class AutoDiffBlock is the most important and most well-documented.
Issues can be reported in the Git issue tracker online at:
https://github.com/OPM/opm-autodiff/issues
To help diagnose build errors, please provide a link to a build log together with the issue description.
You can capture such a log from the build using the `script' utility, e.g.:
LOGFILE=$(date +%Y%m%d-%H%M-)build.log ;
cmake -E cmake_echo_color --cyan --bold "Log file: $LOGFILE" ;
script -q $LOGFILE -c 'cmake ../opm-core -DCMAKE_BUILD_TYPE=Debug' &&
script -q $LOGFILE -a -c 'ionice nice make -j 4 -l 3' ||
cat CMakeCache.txt CMakeFiles/CMake*.log >> $LOGFILE
The resulting file can be uploaded to for instance gist.github.com.