huppd / pintimpact Goto Github PK

Pimpact stands for periodic incompressible Navier--Stokes solvers on massively pallel computers.

To get the code, one has to execute

git clone https://github.com/huppd/PINTimpact.git

in a terminal. This will download the folder PINTimpact into the folder of execution. This folder has the following folder structure

• run
• src
• src_c
• src_f
• test
• XML

The run folder contains various python scripts, that can start multiple jobs and create folder hierarchies for parameter studies or scaling tests. The src folder contains three subfolders, src_c for the C++ part of the code, src_f for the Fortran90 part of the code, and test for the unit-tests. The XML folder contains various xml parameter files. These are used to set up the parameters and problems for the Pimpact solver.

• f90: fortran code
• hpp: c++ header files
• cpp: c++ source files
• cxx: c++ main files
• xml: parameter xml files
• py: python scripts
• txt: text files for cmake

The documentation can be created by executing

doxygen Doxyfile

in the PINTimpact folder. This will generate the documentation in the doc folder. The documentation can be read by opening the file doc/html/index.html by a preferred browser. Further information about compiling and using the code can be found there. This has to be redone, if changes in the code documentation have been done.

We recommend to compile the code in two versions: a release version that is optimized, and a debug version that provides runtime tests and further debugging informations. A user, that is only running the code without adjustments might do without a debug version. A developer, who is working on the performance of the code, might even want to use a third version for profiling.

Pimpact is written partly in C++ and Fortran90, so we recommend to use a compiler that can compile both and directly link those. Test have been done with the intel compiler and the gcc compiler.

Needed libraries are MPI, HDF5, LAPACK, and trilinos.

We are using functionality from the ... Standard so, we recomend using OpenMPI ... or MPICH ....

The recomended HDF5 version is 1.8.13

Depending if your platform already provides an actuall verision of Trilinos (12.1) including the packages Teuchos, Belos, NOX, this step can be skipped and directly proceeded with 2b). A quick installation guide can be found at https://github.com/trilinos/Trilinos/blob/master/INSTALL.rst. A full installation reference can be found at https://trilinos.org/docs/files/TrilinosBuildReference.html. We provide here short customized guide for the needs of Pimpact.

The Trilinos code can be downloaded by

git clone https://github.com/trilinos/Trilinos.git

We reccomend to save the following steps in a bash script for each platform. To install Trilinos after downloading the code, three steps have to be done: make the buildsytem by cmake, build the code by make, install the build by make install. Optionally the build can be tested by ctest.

cmake $ARGS ${TRILINOS_SRC_HOME}

The TRILINOS_SRC_HOME is the path to the downloaded folder. If the git clone comand from above has been executed in the homefolder, then it is TRILINOS_SRC_HOME="~/Trilinos".

-D CMAKE_INSTALL_PREFIX="~/trilinos/debug"

-D CMAKE_BUILD_TYPE:STRING=DEBUG \

-D Trilinos_ENABLE_DEBUG=ON \

-D TPL_ENABLE_MPI:BOOL=ON \

-D Trilinos_ENABLE_TESTS:BOOL=ON \

-D BUILD_SHARED_LIBS:BOOL=OFF \

-D TPL_ENABLE_LAPACK:BOOL=ON \

-D TPL_ENABLE_Boost:BOOL=OFF \

-D TPL_ENABLE_BLAS:BOOL=ON \

-D TPL_ENABLE_HDF5:BOOL=OFF \

-D TPL_ENABLE_MKL:BOOL=OFF \

-D Trilinos_ENABLE_CHECKED_STL:BOOL=OFF \

-D Trilinos_ENABLE_CXX11:BOOL=ON \

-D Trilinos_ENABLE_OpenMP:BOOL=OFF \

-D Trilinos_ENABLE_Anasazi:BOOL=OFF \

-D Trilinos_ENABLE_AztecOO:BOOL=OFF \

-D Trilinos_ENABLE_Fortran:BOOL=ON \

-D Trilinos_ENABLE_Teuchos:BOOL=ON \

-D Trilinos_ENABLE_Kokkos:BOOL=OFF \

-D Trilinos_ENABLE_Belos:BOOL=ON \

-D Trilinos_ENABLE_NOX:BOOL=ON \

-D Trilinos_ENABLE_Epetra:BOOL=OFF \

-D Trilinos_ENABLE_PyTrilinos=OFF \

-D Trilinos_ENABLE_ML:BOOL=OFF \

-D Trilinos_ENABLE_Thyra:BOOL=OFF \

-D Trilinos_ENABLE_Tpetra:BOOL=OFF \

-D Trilinos_ENABLE_Galeri:BOOL=OFF \

After a successful creating the build system, Trilinos can be compiled.

make -j4

Optional: if the Trilinos_ENABLE_TESTS:BOOL=ON has been set the Trilinos build can be tested in the temporary build director by

ctest .

After the succesfull build, Trilinos can be installed by

make install

This will basically move the necessary header and library files to the set CMAKE_INSTALL_PREFIX folder. The temporary Trilions folder can be removed or cleaned afterwards.

We discurage in builts, as cmake is generating plenty of temporal files and folders. We recomend having one or multiple build director in the PINTimpact folder, called accordingly debug, release, or profile.

This comad has to be executed in the build folder for example in ~/PINTimpact/release. The last argument for cmake has to be the path to the main CMakeLists.txt, which can be for example ~/PINTimpact/src

cmake src_path

-DTrilinos_DIR=...

-DfindHDF5=ON/OFF

-DHDF5_INCLUDE_DIRS=...

-DHDF5_LIBRARY_DIRS=...

-DCMAKE_BUILD_TYPE=Release

-DCMAKE_BUILD_TYPE=Debug

-DCMAKE_BUILD_TYPE=Profile

After finishing step 2b) the implementation can be tested. One has to change into the build director and can simply run

ctest .

This will run all test with multiple parameters. The source code for the test can be find in PINTImpact/src/test folder. They are written using the unit testing support provided by the Teuchos package from Trilinos. . The according executable can be found in the build director /test. If the executable are run with a --help flag, they print all possible possible parameter for the command line.

In the folder PINTimpact you can find different executable. The according binarires have the same names just without the suffix .cxx. To print the their comand line parameters they can be executed with the --help flag. This is an overview over the executables

• analyzer3D.cxx analyzes results from peri_navier3D
  • --filename
  • --restart
• convDiff.cxx solver for convection-diffusion problems (has no comandline parameters)
• convDiff2.cxx solver for convection-diffusion problems (has no comandline parameters)
• convDiffMG.cxx solver for convection-diffusion problems (has no comandline parameters)
• convDiffMG2.cxx solver for convection-diffusion problems (has no comandline parameters)
• modeConvDiff.cxx solver for harmonic convection-diffusion problems
  • --filename
  • --realCase
• peri_navier2D.cxx spectral in time solver for two-dimensional Navier--Stokes equations
• peri_navier3D.cxx spectral in time solver for the three-dimesnional Navier--Stokes equations
  • --filename
  • --restart
  • --nf_restart
• peri_navier3DTime.cxx solver with finite differences in time
  • --filename

In the folder *run* are multiple python scripts to run parameter studies.

Depending on the set parameters in the xml file different output files are produced. There are different txt files that contain informations about the performance of the solver such as achived tolerance or number of iterations. Different h5files can be wirtten out, for restarting, analyzing, or visualizing. If the fields are not written for restarting they are interpolated on the pressure grid and are accompanied with xmf files. These xmf files can be opened with for example paraview or VisIt.