Download and Compilation¶
To get limbo, simply clone the source code from https://github.com/resibots/limbo with git, or download it as a zip.
Optional but highly recommended¶
Intel TBB is not mandatory, but highly recommended; TBB is used in Limbo to take advantage of multicore architectures.
NLOpt [mirror: http://members.loria.fr/JBMouret/mirrors/nlopt-2.4.2.tar.gz] with C++ binding:
./configure --with-cxx --enable-shared --without-python --without-matlab --without-octave sudo make install
The Debian/Unbuntu NLOpt package does NOT come with C++ bindings. Therefore you need to compile NLOpt yourself. The brew package (OSX) comes with C++ bindings (brew install homebrew/science/nlopt).
libcmaes. We advise you to use our own fork of libcmaes (branch fix_flags_native). Make sure that you install with sudo or configure the LD_LIBRARY_PATH accordingly. Be careful that gtest (which is a dependency of libcmaes) needs to be manually compiled even if you install it with your package manager (e.g. apt-get):
sudo apt-get install libgtest-dev sudo cd /usr/src/gtest sudo mkdir build && cd build sudo cmake .. sudo make sudo cp *.a /usr/lib
Follow the instructions below (you can also have a look here):
git clone https://github.com/resibots/libcmaes.git cd libcmaes git checkout fix_flags_native
Configuring with Makefiles:
./autogen.sh ./configure make -j4
mkdir build cd build cmake .. make -j4
In addition, you should be careful to configure libcmaes to use the same Eigen3 version as what you intend to use with Limbo (configuring with Makefiles):
or (configuring with CMake):
cmake -DEIGEN3_INCLUDE_DIR=YOUR_DESIRED_DIR/include/eigen3 ..
Additionally, you can enable the usage of TBB for parallelization (configuring with Makefiles):
or (configuring with CMake):
cmake -DUSE_TBB=ON -DUSE_OPENMP=OFF ..
- Intel MKL is supported as backend for Eigen. In our experience, it provided best results when compiling with Intel’s Compiler (ICC)
- LAPACKE/BLAS is supported as a backend for Eigen (version>=3.3). In our experience, it gives high speed-ups with big matrices (i.e., more than 1200 dimensions) and hurts a bit the performance with small matrices (i.e., less than 800 dimensions). You can enable LAPACKE/BLAS by using the
--lapacke_blasoption (if you have Eigen3.3 or later).
- Sferes2 if you plan to use the multi-objective bayesian optimization algorithms (experimental).
Like most build systems, it has a configuration and build steps, described bellow.
Make sure that the waf file has execution rights.
The first step is to configure your waf environment. For this, assuming that you are in the root directory of Limbo, you have to run the command:
If everything is okay, you should expect an output like this:
Setting top to : /path/to/limbo Setting out to : /path/to/limbo/build Checking for 'g++' (c++ compiler) : /usr/bin/g++ Checking for 'gcc' (c compiler) : /usr/bin/gcc Checking boost includes : 1_55 Checking boost libs : ok Checking Intel TBB includes : not found Checking for compiler option to support OpenMP : -fopenmp Checking Intel MKL includes : not found ['-Wall', '-std=c++11', '-O3', '-march=native', '-g']
The actual ouput may differ, depending on your configuration and installed libraries.
Waf should automatically detect the libraries if they where installed in the default folders, but if it doesn’t, you can use the following command-line options to indicate where they are:
--boost-libs /path/to/boost-libraries[.a, .so, .dynlib]
Note that Sferes2 won’t be used unless you specify it’s installation folder.
You can also specify a different compiler than the default, setting the environment variables
A full example:
CC=icc CXX=icpc ./waf configure --sferes ~/sferes2 --mkl ~/intel/mkl --tbb ~/intel/tbb
The second step is to run the build command:
Depending on your compiler, there may be some warnings, but the output should end with the following lines:
'build' finished successfully (time in sec)