Mini-MDFT

The Molecular Density Functional Theory is a collaborative work led by Daniel Borgis and Maximilien Levesque:

Using Mini-MDFT

First of all, you need to clone this repo.

git clone --recursive https://github.com/yasahi-hpc/Mini_MDFT.git

Prerequisites

Mini-MDFT is developed with C++20, thus you need newer compilers. To use Mini-MDFT, we need the followings:

• CMake 3.22+
• Kokkos 4.4+
• gcc 11.0.0+ (CPUs, c++20 support is necessary)
• IntelLLVM 2025.0.0+ (CPUs, Intel GPUs)
• nvcc 11.0.0+ (NVIDIA GPUs)
• rocm 5.6.0+ (AMD GPUs)

Compile and run

For compilation, we basically rely on the CMake options for Kokkos. For example, the compile options for A100 GPU is as follows.

cmake -B build \
      -DCMAKE_CXX_COMPILER=g++ \
      -DCMAKE_BUILD_TYPE=Release \
      -DCMAKE_CXX_STANDARD=20 \
      -DKokkos_ENABLE_CUDA=ON \
      -DKokkos_ARCH_AMPERE80=ON \
      -DKokkosFFT_ENABLE_HOST_AND_DEVICE=ON \
      -DMini_MDFT_ENABLE_TESTS=ON \
      -DMini_MDFT_ENABLE_INTERNAL_KOKKOS=ON \
      -DMini_MDFT_ENABLE_INTERNAL_KOKKOSFFT=ON
cmake --build build -j 8

To run the tests, please run the following command.

cd build
ctest --output-on-failure

To run the code, please run the following command.

cd build
./src/mini-MDFT -filename ../input/dft2.json -solute ../input/solute.json -luc_file ../data/dcf/tip3p/tip3p-ck_nonzero_nmax3_ml

Benchmark tests have been implemented using Google Benchmark. In order to build the benchmark tests add the following flag when building the project:

-DMini_MDFT_ENABLE_BENCHMARKS=ON

Google Benchmark will be automatically downloaded and built if it is not already installed in your system. In order to run the tests, go to the directory perf_test inside the build directory and run the executables, whose names will be of the form Kokkos_Benchmark_*. e.g., Kokko_Benchmark_Solvent.

Development strategy

1. Port Mini-MDFT function by function with unit-testing
2. Add docstrings for each function
3. Reconstruct I/O with json/yaml and hdf5/netcdf
4. Create a minimal benchmark case for integrated testing and compare the result with Fortran
5. Making a PR for each modification and ask for a review if you wish
6. Formatting your files by clang-format -i <file_you_have_modified>
7. Feel free to use AI tools to convert Fortran code into C++, adding unit-test and docstrings, etc

LICENSE

Mini-MDFT is distributed under the MIT license.