Compiling RuNNer
RuNNer requires LAPACK libraries, which can be downloaded for free e.g. from www.netlib.org. After downloading and compiling LAPACK, please proceed as follows:
Compilation with make
Friendly reminder of the setting for ifort
Please refer to FAQ 7.1.9
in the documentation.
-
Go to the source folder and edit the Makefile by inserting your compiler, e.g.
ifort
orgfortran
, (g95
should not be used), your compiler options and the library path to LAPACK (e.g. themkl
path or the path of the downloaded netlib libraries). For the mpi version ofRuNNer
also some MPI settings are required (only mode 3 ofRuNNer
is currently parallelized in a meaningful way). The serial version should be the first to try. The fitting part of RuNNer is currently parallelized by using openmp, just add-f openmp
in FFLAGS of RuNNer Makefile and compile with serial version. Before executing the executable ofRuNNer
, type the commandexport OMP_NUM_THREADS=N
for N cores. Since the openmp parallelization is for the symmetry function calculation only, a speed-up factor of 2 can be expected at most. -
For compiling the serial version of
RuNNer
please typemake serial
You should obtain a binary called
RuNNer.serial.x
. -
For compiling the parallel version of
RuNNer
please typemake mpi
You should obtain a binary called
RuNNer.mpi.x
.
Compilation using CMake
For the version 2.0 or above, RuNNer
uses CMake for the compilation. The
procedure is the following:
- Create the build subdirectory in the RuNNer main directory:
mkdir build
- Move to the build directory and run CMake:
cd build cmake .. -DOPENMP=ON -DINTEL=ON -DDEBUG=OFF
- run make:
make
CMake takes three parameter which can be either set to ON
or OFF
:
- OPENMP: Enables OpenMP parallelization.
- INTEL Uses ifort to compile RuNNer. If disabled, gfortran is used.
- DEBUG Compile with debugging flags. Do not forget to disable this for production runs, since performance is significantly reduced.
Now RuNNer
is ready for use.