Skip to content

Latest commit

 

History

History
112 lines (88 loc) · 6.49 KB

README.md

File metadata and controls

112 lines (88 loc) · 6.49 KB
Raw

DGSWEM V2

CircleCI

Getting Started

dgswem-v2 has a considerable number of dependencies. If you are simply interested in giving dgswemv2 a quick test. We recommend one of two options. First, if one does not want to build all of the dependencies outlined below, running

cmake -DCMAKE_PREFIX_PATH=<YAML_CPP_INSTALL_PATH>\
      -DBUILD_EXAMPLES=On\
    <DGSWEMV2_ROOT>

will build the code to run using a Runge-Kutta discontinuous Galerkin discretization in serial along with the manufactured solution.

Another alternative is to use our Docker image, which contains the necessary dependencies to build all dgswemv2 targets. Simply, run

docker pull bremerm31/dgswemv2:latest
docker run -it bremerm31/dgswemv2

The docker container contains all dependencies, and builds all targets.

For tips, on how to get started, we recommend looking at our users guide, which can be found in documentation/users-guide/dgswem-v2-users-guide.pdf. Basic examples about how to run the code and the design philosophy behind dgswemv2 can be found here.

Lastly, for any further questions, we encourage you to either open an issue on the github repository or join our slack channel. Simply open an issue in the repo with the Slack tag and the email address with which you would like to be added.

Building instructions

Dependencies

In designing any software package there are clear benefits and disadvantages to using libaries. In dgswem-v2, we have the following dependencies:

Dependency Function Dependent Targets
yaml-cpp YAML file parser All targets
Metis Graph partitioner partitioner
HPX Asynchronous Runtime *_SWE_HPX
OpenMP Application threading *_SWE_OMPI
MPI Message passing *_SWE_OMPI,*_SWE__HPX
Eigen Linear Algebra Library EHDG_*

To begin select a work directory. On the TACC machines we recommend using the $WORK directory and copying all of the meshes into $SCRATCH for running jobs. yaml-cpp, Metis, and HPX are all libraries. To assist the user in installing the Libraries we have added some bash scripts in scripts/build to assist the user in compiling dgswem-v2.

To begin the installation process, you will need to set up a configuration file. From the root of the repository, go to scripts/build. Open up config.txt and define a machine according to your preferences. Note the build-XXXX.sh scripts will use config.txt by default. However, one can also use custom configuration file as follows:

    ./build-XXXX.sh -c custom_config.txt

Note that for HPX, the library's build-type must be consistent with the applications build-type.

Installing yaml-cpp

From $WORK (as defined in the configuration file)

    git clone [email protected]:jbeder/yaml-cpp.git
    cd /path/to/build/scripts/
    ./build-yaml-cpp.sh

Installing hpx

The building of hpx requires a lot of dependencies. In particular, for our bash script, we require that jemalloc, boost, hwloc, and an MPI implementation be installed. Note that most clusters typically come installed with boost, hwloc, and an MPI implementation. Thus, if you have already have installed versions of the afore mentioned libraries, skip the relevant build scripts.

    cd $WORK
    git clone https://github.com/STEllAR-GROUP/hpx
    cd /path/to/build/scripts
    ./build-hwloc.sh
    ./build-boost.sh
    ./build-jemalloc.sh
    ./build-hpx.sh

Installing eigen

To build the hybridized discontinuous Galerkin targets, the application requires Eigen3. Although it is a header only library, we have provided a build-eigen.sh script, which will run download Eigen 3.3.4 from the internet, run cmake to run various checks, and lastly, install the headers to the config.txt specified location.

cd /path/to/build/scripts
./build-eigen.txt

Building the Application

Assuming that ${INSTALL_PATH} is the path defined in the configuration file. The dgswemv2 uses an out of source build. The application can be built as follows:

    cd /path/to/dgswemv2/
    mkdir build
    cmake -DCMAKE_BUILD_TYPE=<build type> -DCMAKE_PREFIX_PATH=${INSTALL_PATH} ..

Note that there are some additional options, which will create additional targets. These typically require additional dependencies.

CMake Option Description
USE_OMPI Enables MPI OpenMP parallelization; builds target DG_HYPER_SWE_OMPI
USE_HPX Enables HPX parallelization; builds target DG_HYPER_SWE_HPX
COMPILER_WARNINGS Display compiler warnings
SET_VERBOSE Set the makefile compilation output to Verbose
BUILD_EXAMPLES Build additional executables to run the examples
RKDG_SWE Build Runge-Kutta discontinuous Galerkin targets
EHDG_SWE Build explicit Hybridized discontinuous Galerkin targets
IHDG_SWE Build implicit hybridized discontinuous Galerkin targets

Note that by default RKDG_SWE is set to On and associated targets will be built by cmake.

License

DGSWEM V2 is licensed under the MIT license. The following files have been copied (and potentially modified) from other repositories. Their licenses are inlined within the files:

  • cmake/modules/FindMETIS.cmake
  • documentation/doxygen-bootstrapped/customdoxygen.css
  • documentation/doxygen-bootstrapped/doxy-boot.js
  • documentation/doxygen-bootstrapped/footer.html
  • documentation/doxygen-bootstrapped/header.html
  • source/utilities/linear_algebra/serialization/blaze_vector.hpp
  • source/utilities/linear_algebra/serialization/blaze_matrix.hpp

N.B. The files in documentation/doxygen-boostrapped are taken from feature/support-doxygen-1.1.12+ branch of the Velron/doxygen-bootstrapped repository. The copy of the Apache-2.0 licesnse can be found in documentation/doxygen-bootstrapped/APACHE_2.0_LICENSE of this repository or in the original repository.