Installation

This package uses a Multiple stream Mersenne Twister PRNG to generated the required random numbers. This package uses FLAP to deal with comand line arguments. The FLAP pacakge is recommended to install using the FoBiS.py, which is a building system for Fortran projects. Currently, the code can only be run on CPUs.

Requirments

  • Programs that are commonly installed on HPC clusters

    • Git: for cloning the code from GitHub

    • CMake: 3.9.0 or higher

    • GCC or Intel Fortran compilers.

    • MPI: OpenMPI or MPICH. It should work with the others.

    • HDF5: parallel version.

    Tip

    On Perlmutter@NERSC, you can load the modules by module load cpu cmake cray-hdf5-parallel

  • FoBiS.py: a building system for Fortran projects

    • The installation wiki: https://github.com/szaghi/FoBiS/wiki/Install.

    • It is recommended to use PyPI to install it: pip install FoBiS.py --user

    • After installing, make sure that FoBiS.py is in your $PATH by including

      export PATH=$HOME/.local/bin/:$PATH

      in your .bashrc.

  • Multiple stream Mersenne Twister PRNG: for generating pseudorandom numbers parallelly.

    This program module splits single long period random number series from Mersenne Twister (MT) into multiple (almost) independent streams.

    • Download: http://theo.phys.sci.hiroshima-u.ac.jp/~ishikawa/PRNG/mt_stream_f90.tar.gz

    • Load the same packages as listed above.

      tar zxvf mt_stream_f90.tar.gz
cd mt_stream_f90-1.11

      Depending on the system type, you might need to modify FC in Makefile by hand. The default is ifort, which is for the Intel compilers. For gfortran, please use the script config/gfortran_mt_stream.sh to make the relevant changes in Makefile. First, copy the script to mt_stream_f90-1.11. Then,

      ./gfortran_mt_stream.sh

      Note

      If you are using newer versions of gfortran, there will be compiling errors. Therefore, we need to update the source code slightly using config/fix_mt_stream.sh. First, copy the script to mt_stream_f90-1.11. Then,

      ./fix_mt_stream.sh

    • You can then compile the code by

      make

    • After compiling the code, you need to set the environment variable MT_STREAM to the installation directory. For example,

      export MT_STREAM=$HOME/local/mt_stream_f90-1.11

      This environment variable will be used to compile the GPAT code.

      Tip

      You can put this into your .bashrc.

Download

git clone https://github.com/xiaocanli/stochastic-parker

Install

In the directory stochastic-parker,

mkdir build
cd build
cmake ..
make
make install

To turn on OpenMP parallelization, please use cmake -DUSE_OPENMP="On" ... To turn on AVX512 for the KNL nodes, please also include -DUSE_AVX512="On".

Tip

On Perlmutter@NERSC, you can load the modules by module load cpu cmake cray-hdf5-parallel python

Note

Please use the same version of python as when installing FoBiS.py. Otherwise, FLAP will not be compiled correctly.

Note

The code is not carefully optimized for the KNL nodes. Please use it in caution.

You can create a soft link of the executable stochastic-mhd.exec in the scratch filesystem for running the code.