Skip to content

Latest commit

 

History

History
106 lines (76 loc) · 3.5 KB

install.md

File metadata and controls

106 lines (76 loc) · 3.5 KB

Build and install

We suggest two different ways to build or install AmgXWrapper.

1. Compile the source code together with users' applications

Since AmgXWrapper doesn't have a lot of code or complex hierarchy, it's very easy to include the source code in the building process of your applications. The poisson and solveFromFiles in the folder example belong to this use case.

2. Build and install AmgXWrapper with CMake

AmgXWrapper can be compiled and built as a library and can be installed to a user's preferred path. We provide CMakeLists.txt for this purpose. Our CFD solver, PetIBM is an example of using AmgXWrapper as a library.

Step 1

At any location you like, make a temporary folder to held compiled files and then go into that folder. For example,

$ mkdir ${HOME}/build
$ cd ${HOME}/build

Step 2

Run CMake.

$ cmake \
    -D CMAKE_INSTALL_PREFIX=${PATH_TO_WHERE_YOU_WANT_TO_INSTALL_AmgXWrapper} \
    -D PETSC_DIR=${PATH_TO_PETSC} \
    -D PETSC_ARCH=${THE_BUILD_OF_PETSC_YOU_WANT_TO_USE} \
    -D CUDA_DIR=${PATH_TO_CUDA} \
    -D AMGX_DIR=${PATH_TO_AMGX} \
    ${PATH_TO_AmgXWrapper_SOURCE}

Other optional CMake parameters are:

  • CUDA_HOST_COMPILER: the underlying C compiler for CUDA compiler nvcc. If the default C compiler is too new for CUDA, users may need to use this parameter to choose an older version of C compiler.

  • CMAKE_C_COMPILER: the default is mpicc. This is the C compiler used to compile C codes in AmgXWrapper. If a user wants to change the underlying C compiler used by MPI compiler, he/she should instead use the environment variable OMPI_MPICC for this purpose. See Compiling MPI applications.

  • CMAKE_CXX_COMPILER: the default is mpicc. This is the C++ compiler used to compile C++ codes in AmgXWrapper. If a user wants to change the underlying C++ compiler used by MPI compiler, he/she should instead use the environment variable OMPI_MPICXX for this purpose. See Compiling MPI applications.

  • CMAKE_BUILD_TYPE: the default is RELEASE. RELEASE mode will use the flags -O3 -DNDEBUG. Another option is DEBUG, which is using flag -g.

  • CMAKE_CXX_FLAGS and CMAKE_C_FLAGS: set customized flags here if desired.

  • BUILD_SHARED_LIBS: the default is ON. This will create a shared library libAmgXWrapper.so. To create a static library (libAmgXWrapper.a) only, set this argument to OFF.

Step 3

Build the library.

$ make -j <number of threads used to compiling>

Step 4 (optional)

If Doxygen exists, and if CMake is able to find Doxygen, the API documentation can be built with

$ make doc

This will build a HTML version of API documentation. The homepage is at doc/html/index.html under the build directory before installation.

Step 5

Install the library, header, and documents to your preferred location.

$ make install

After installation, the build directory can be safely deleted. For example, here we can delete ${HOME}/build.

Under the installation location, lib (or lib64, depends on the system) contains the library, libAmgXWrapper.so or libAmgXWrapper.a. The header file AmgXSolver.hpp will be in the folder include, and the documents (if available) will be in the folder doc.