Requirements
The library can be compiled on Linux and OS X systems with GNU, Intel, or Clang compilers. For OS X the recommended toolchain is GNU or Clang installed with MacPorts. AppleClang si partially supported only with an external OpenMP installation.
Two branches are available in the project. The master branch is supposed to be stable (as stable as a research project can be), while develop inlcudes the newest enhancements before they are ready to be merged in master.
Cloning the repository
The repository and its submodules can be cloned by
git clone [email protected]:zap150/besthea.git
cd besthea
git submodule update --init --recursive
Pulling updates
When pulling updates one should also pull the possibly updated submodules as
git pull --recurse-submodules
git submodule update --init --recursive
Dependencies
Beside the Boost, Eigen, and Lyra submodules, BESTHEA requires the installation of MPI and Intel MKL. To configure the project make sure that the MKLROOT and LIBRARY_PATH variables are set by calling the scripts mklvars.sh and compilervars.sh provided by the MKL installation (in the case of Intel OneAPI, use the script setvars.sh).
OpenMP is a dependency usually accompanying a compiler. For a note on AppleClang see the next section.
If one wants to use the GPU-accelerated functionality implemented in the besthea_cuda library, CUDA should also be installed.
Build
The compilation of BESTHEA is based on CMake. Since in-source build is disabled, create a build directory and call cmake from within as
mkdir build
cd build
cmake ..
make
make install
To specify the compiler you can either use environment variables as
CC=icc CXX=icpc cmake ..
or CMake variables
cmake -DCMAKE_C_COMPILER=icc -DCMAKE_CXX_COMPILER=icpc ..
The project includes example executables testing the library. These can be disabled by setting
cmake -DBUILD_EXAMPLES=OFF ..
AppleClang does not come with a native support of OpenMP. This can be installed e.g. via MacPorts. As of now, an include hint for omp.h is necessary, e.g.
CXXFLAGS="-isystem /opt/local/include/libomp"
The BESTHEA library uses OpenMP SIMD. To fully leverage its potential we recommend the Intel compiler with optimisation flags set in accordance with the system. E.g. on a Skylake CPU we would use
CXXFLAGS="-xcore-avx512 -qopt-zmm-usage=high" cmake ..
Vectorisation with GNU or Clang does not work optimally from our experience, however, one can try to use CXXFLAGS="-march=skylake-avx512", CXXFLAGS="-mavx512*", and similar.
To specify the vector length processed in OpenMP SIMD loops, i.e. the simdlen clause, we provide the variable BESTHEA_SIMD_WIDTH implicitly set to 8. This can be modified by e.g.
cmake -DBESTHEA_SIMD_WIDTH=4 ..
If CUDA is found on the system and CMake >= 1.18 is used, the library besthea_cuda containing the GPU-accelerated functionality is also built. To alter this behaviour, set the value of the BESTHEA_CUDA variable to enable, disable or auto, e.g.
cmake -DBESTHEA_CUDA=disable ..
If the value is enable and any of the prerequisities are not met, the configuration step fails with an error. disable entirely disables compilation of the GPU-accelerated code. auto is the default (equivalent to not setting the value of BESTHEA_CUDA at all), which tries to build the besthea_cuda library if it is possible, and if not, no errors are produced.
Usage
The install target installs the static libraries to ${CMAKE_INSTALL_PREFIX}/lib/besthea, include files to ${CMAKE_INSTALL_PREFIX}/include/besthea, and executable examples to ${CMAKE_INSTALL_PREFIX}/bin/besthea together with example mesh files. One can run an example
./uniform_tensor_neumann --mesh cube_192.txt --grid grid_cube_xy.txt
or
./uniform_tensor_neumann --help
to see all command line options.
Please cite this research as:
-
J. Zapletal, R. Watschinger, G. Of, M. Merta, Semi-analytic integration for a parallel space-time boundary element method modeling the heat equation, Comput. Math. Appl. 103 (2021) 156-170, 10.1016/j.camwa.2021.10.025.
-
R. Watschinger, M. Merta, G. Of, J. Zapletal, A parallel fast multipole method for a space-time boundary element method for the heat equation, SIAM J. Sci. Comput. Vol. 44, Iss. 4 (2022), pp. C320-C345, https://doi.org/10.1137/21M1430157.
See this guide to reproduce experiments from the paper. -
R. Watschinger, G. Of, A Time-Adaptive Space-Time FMM for the Heat Equation, Comput. Methods Appl. Math., 2022. https://doi.org/10.1515/cmam-2022-0117.
See this guide to reproduce experiments from the paper.
Other related publications:
- R. Watschinger, G. Of, An integration by parts formula for the bilinear form of the hypersingular boundary integral operator for the transient heat equation in three spatial dimensions, J. Integral Equ. Appl. Vol. 34, Iss. 1 (2022), pp. 103-133, https://doi.org/10.1216/jie.2022.34.103.
See the project website.
Authors acknowledge the support provided by the Czech Science Foundation under the project 19-29698L, the Austrian Science Fund (FWF) under the project I 4033-N32, and by The Ministry of Education, Youth and Sports from the Large Infrastructures for Research, Experimental Development, and Innovations project 'e-INFRA CZ – LM2018140'.