CMake

CP2K can be compiled with the CMake build system. CMake is used to detect CP2K dependencies and configure the compilation process.

CP2K dependencies should be installed independently, either manually or using a package manager. See [CP2K’s Spack Documentation] for documentation on how to use the [Spack] package manager to build CP2K’s dependencies.

Using CMake

The most common CMake workflow is to create a build/ directory in the root directory of the source tree, and run CMake as follows:

cd <CP2K_REPOSITORY>
mkdir build/
cmake -S . -B build             # -S <SOURCE_DIR> -B <BUILD_DIR>
cmake --build build -- -j 32    # -build <BUILD_DIR> -- <BUILD_TOOL_OPTIONS>

<CP2K_REPOSITORY> is a placeholder for the path to the CP2K repository, containing the source code to be compiled.

Generators

By default, CMake generates GNU Makefiles (on Linux). With CMake, it is possible to generate files for other build systems such as [Ninja]:

cmake -S . -B build -GNinja

Release and Debug Builds

CMake allows to specify a build type. -DCMAKE_BUILD_TYPE=Release turns on optimizations, while -DCMAKE_BUILD_TYPE=Debug turns on debug options.

Requirements

The minimum requirements to use the CMake build system are the following:

  • CMake

  • C, C++, and Fortran compilers

  • [DBCSR]

  • OpenMP]

  • BLAS and LAPACK

For an MPI build, the following dependencies are also required:

  • MPI

  • ScaLAPACK

DBCSR

[DBCSR] is a required dependency of CP2K, and CMake expects [DBCSR] as an external dependency.

BLAS, LAPACK, and ScaLAPACK

The major vendors’ implementations of BLAS and LAPACK are supported. If CP2K is compiled with MPI support (-DCP2K_USE_MPI=ON), ScaLAPACK is also required.

-DCP2K_BLAS_VENDOR and -DCP2K_SCALAPACK_VENDOR can be used to control which vendor library to use.

To build with Intel OneAPI MKL:

cmake -S . -B build -DCP2K_BLAS_VENDOR=MKL -DCP2K_SCALAPACK_VENDOR=MKL
cmake --build build

To build with Cray LibSci:

cmake -S . -B build -DCP2K_BLAS_VENDOR=SCI -DCP2K_SCALAPACK_VENDOR=SCI
cmake --build build

Build Customization

Default Dependencies

All optional dependencies are turned off by default, with the exception of MPI.

For simplicity, -DCP2K_BUILD_OPTIONS is provided to turn on some of the optional dependencies. -DCP2K_BUILD_OPTIONS can take the following values:

  • CUSTOM (default)

  • DEFAULT

  • MINIMAL

  • FULL

  • SERIAL

By default, -DCP2K_BUILD_OPTIONS=CUSTOM, meaning that each dependency must be turned on explicitly with -DCP2K_USE_<LIBRARY>=ON. <LIBRARY> is the name of the optional dependency (for example: -DCP2K_USE_COSMA=ON, -DCP2K_USE_LIBXC=ON, …).

Please refer to the CMakeLists.txt file for an up-to-date list of the dependencies enabled by each option.

GPUs

CP2K is GPU-accelerated. In order to enable GPU acceleration with [CUDA] or [HIP], -DCP2K_USE_ACCEL can be used:

  • -DCP2K_USE_ACCEL=CUDA: enables [CUDA]

  • -DCP2K_USE_ACCEL=HIP: enables [HIP]

The target architecture can be selected with -DCP2K_WITH_GPU.

CUDA can either be build using the NVIDIA HPC package (-DCP2K_USE_NVHPC=ON) or the CUDA toolkit package(-DCP2K_USE_NVHPC=OFF).

cuSOLVERMp

cuSOLVERMp is activated with -DCP2K_USE_CUSOLVER_MP=ON has two dependicies:

  • CAL: requires libcal.* in the Path

  • UCC: requires libucc.* libucs.* in the Path

Example

Build CP2K with CUDA acceleration for Nvidia A100 GPUs, with multiple optional dependencies:

cd <CP2K_REPOSITORY> && make build/
cmake -S . -B build \
    -GNinja \
    -DCP2K_USE_LIBXC=ON \
    -DCP2K_USE_LIBINT2=ON \
    -DCP2K_USE_SPGLIB=ON \
    -DCP2K_USE_ELPA=ON \
    -DCP2K_USE_SPLA=ON \
    -DCP2K_USE_SIRIUS=ON \
    -DCP2K_USE_COSMA=ON \
    -DCP2K_USE_ACCEL=CUDA -DCP2K_WITH_GPU=A100

cmake --build build -j 32

[CP2K’s Spack Documentation]: [Spack]: https://spack.readthedocs.io/en/latest/ [DBCSR]: https://cp2k.github.io/dbcsr/develop/ [CUDA]: https://developer.nvidia.com/cuda-toolkit [HIP]: https://rocm.docs.amd.com/projects/HIP/en/latest/