-
Notifications
You must be signed in to change notification settings - Fork 28
CMake options
CMake options can either be specified with the -D flag:
$ cmake /path/to/source -D<variable1>=<value1> -D<variable2>=<value2> ...
or using the ccmake interface, which is launched in the exisiting build directory (after cmake was called) with:
$ ccmake .
| Option | Default value | Description |
|---|---|---|
| CMAKE_BUILD_TYPE | The build type. Options are: [empty], Debug, Release, RelWithDebInfo, MinSizeRel. |
According to the build type CMake sets different optimization and debug compiler flags.
Release is recommended for production runs.
| Option | Default value | Description |
|---|---|---|
| FFTW_INCLUDE_DIR | Path to fftw3.h. | |
| FFTW_LIBRARY | The FFTW3(-compatible) library. |
If the -D flag is used, multiple libraries have to be separated by semicolons and enclosed by double quotes:
-DFFTW_LIBRARY="/path/to/lib1.a;/path/to/lib2.a"
If you use a compiler wrapper that automatically links against BLAS and LAPACK libraries, you can prevent CMake from searching for these libraries by adding the following option to the cmake command:
-DDCA_HAVE_LAPACK=TRUE
| Option | Default value | Description |
|---|---|---|
| DCA_WITH_MPI | ON | Enable MPI. |
If MPI is enabled, CMAKE_CXX_COMPILER must be set to the MPI compiler wrapper.
| Option | Default value | Description |
|---|---|---|
| DCA_BUILD_DCA | ON | Build main_dca.cpp. |
| DCA_BUILD_ANALYSIS | ON | Build main_analysis.cpp. |
| DCA_BUILD_CLUSTER_SOLVER_CHECK | OFF | Build cluster_solver_check.cpp. |
| Option | Default value | Description |
|---|---|---|
| DCA_WITH_TESTS_FAST | OFF | Build DCA++'s fast tests. |
| DCA_WITH_TESTS_EXTENSIVE | OFF | Build DCA++'s extensive tests. |
| DCA_WITH_TESTS_PERFORMANCE | OFF | Build DCA++'s performance tests. (Only in Release mode.) |
| TEST_RUNNER | Command for executing (MPI) programs. Required if MPI is enabled or on systems that use a command for launching executables (e.g. aprun or srun). | |
| MPIEXEC_NUMPROC_FLAG | -n | Flag used by TEST_RUNNER to specify the number of processes. |
| MPIEXEC_PREFLAGS | Flags to pass to TEST_RUNNER directly before the executable to run. |
| Option | Default value | Description |
|---|---|---|
| DCA_WITH_CUDA | OFF | Enable GPU support. |
| DCA_WITH_PINNED_HOST_MEMORY | OFF | Enable pinned host memory. (advanced) |
| CUDA_GPU_ARCH | sm_60 | Name of the real architecture to build for. |
| CUDA_TOOLKIT_ROOT_DIR | Path to the CUDA Toolkit. Determined by CMake. Set it manually, if CMake cannot find it. |
|
| MAGMA_DIR | Path to the MAGMA installation directory. Hint for CMake to find MAGMA. |
In some cases it is necessary to set the environment variable CUDA_LIB_PATH in order for CMake to find CUDA.
| Option | Default value | Description |
|---|---|---|
| DCA_CLUSTER_SOLVER | CT-AUX | The cluster solver for the DCA(+) loop. Options are: CT-AUX, SS-CT-HYB. |
| DCA_WITH_THREADED_SOLVER | ON | Use multiple walker and accumulator threads in the cluster solver. |
| DCA_LATTICE | square | Lattice type, options are: bilayer, square, triangular, twoband_chain, singleband_chain. |
| DCA_POINT_GROUP | D4 | Point group symmetry, options are: C6, D4. |
| DCA_MODEL | tight-binding | Model type, options are: tight-binding. |
| DCA_RNG | std::mt19937_64 | Random number generator, options are: std::mt19937_64, std::ranlux48, custom. See Random number generator for details. |
| DCA_RNG_CLASS | Class name including namespaces of the custom random number generator. | |
| DCA_RNG_HEADER | Header file of the custom random number generator. | |
| DCA_RNG_LIBRARY | Custom random number generator library. | |
| DCA_PROFILER | None | Profiler type, options are: None, Counting, PAPI. |
| DCA_WITH_AUTOTUNING | OFF | Enable auto-tuning. Needs a profiler type other than 'None'. (advanced) |
| DCA_WITH_GNUPLOT | OFF | Enable Gnuplot. |
| DCA_WITH_SINGLE_PRECISION _MEASUREMENTS |
OFF | Measure in single precision. |
| DCA_WITH_SINGLE_PRECISION _COARSEGRAINING |
OFF | Coarsegrain in single precision. (advanced) |
| DCA_WITH_QMC_BIT | OFF | Enable QMC solver built-in tests. (advanced) |
All examples assume the directory tree:
|____build (current working directory)
|____dca_source
-
Builds main_dca, main_analysis (default) and the fast tests. MPI is disabled.
$ cmake ../dca_source \ -DFFTW_INCLUDE_DIR=/opt/local/include \ -DFFTW_LIBRARY=/opt/local/lib/libfftw3.a \ -DDCA_WITH_MPI=OFF \ -DDCA_WITH_TESTS_FAST=ON -
Builds main_dca, main_analysis (default) and both fast and extensive tests with MPI enabled (default) and debug flags. TEST_RUNNER is set to the mpirun command corresponding to the MPI compiler wrapper, that has been assigned to CXX in the first line.
$ CXX=/opt/local/bin/mpicxx-mpich-clang36 cmake ../dca_source \ -DCMAKE_BUILD_TYPE=Debug \ -DFFTW_INCLUDE_DIR=/opt/local/include \ -DFFTW_LIBRARY=/opt/local/lib/libfftw3.a \ -DDCA_WITH_TESTS_FAST=ON \ -DDCA_WITH_TESTS_EXTENSIVE=ON \ -DTEST_RUNNER=/opt/local/bin/mpirun-mpich-clang36 -
CSCS Piz Daint [Cray XC40 (multi-core)]
Builds main_dca and main_analysis (default) with MPI enabled (default) and optimization flags. BLAS and LAPACK are automatically linked by the compiler wrapper CC. TEST_RUNNER is set to the system's command to launch applications, srun.$ cmake ../dca_source/ \ -DCMAKE_BUILD_TYPE=Release \ -DDCA_HAVE_LAPACK=TRUE \ -DTEST_RUNNER=srun -
CSCS Piz Daint [Cray XC50 (hybrid CPU-GPU)]
Builds main_dca and main_analysis (default) with MPI (default) and GPU support enabled and optimization flags. The path to the CUDA toolkit is set and the directory to the MAGMA installation directory speficied (MAGMA is assumed to be installed with EasyBuild). BLAS and LAPACK are automatically linked by the compiler wrapper CC. TEST_RUNNER is set to the system's command to launch applications, srun.$ cmake ../dca_source/ \ -DCMAKE_BUILD_TYPE=Release \ -DDCA_HAVE_LAPACK=TRUE \ -DTEST_RUNNER=srun \ -DDCA_WITH_CUDA=ON \ -DCUDA_TOOLKIT_ROOT_DIR=/opt/nvidia/cudatoolkit8.0/8.0.44_GA_2.2.7_g4a6c213-2.1 \ -DMAGMA_DIR=$EBROOTMAGMA