Skip to content

Latest commit

 

History

History
98 lines (76 loc) · 6.01 KB

usepetibmapi.md

File metadata and controls

98 lines (76 loc) · 6.01 KB
Raw

Use PetIBM API

In the installation path, PetIBM also provides a library and header files that have all necessary API to build a flow solver.

API Documentation

The API documentation can be found in the Modules section of the Doxygen pages. The subsection of PetIBM building blocks shows details of the elements for building a flow solver. The subsection of Flow solvers and utilities shows the flow solvers built on those building blocks.

A local version of Doxygen pages can be created with the doxygen utility. To do that, go into the doc folder of PetIBM, and run the doxygen command. HTML pages will be created in the subfolder html of the doc directory. Users can, then, open the main page of the manual, html/index.html, using their preferred browser.

API usage examples

The source code of the flow solvers mentioned in the API documentation may be unfriendly for beginners, so we provide other simplified solvers in the folder $PETIBM_DIR/examples/api_examples. For example, liddrivencavity2d provides step-by-step instructions in the comments of the source code. (See the README files in those folders for additional details.)

Compiling and linking against PetIBM

Header files

Let $PETIBM_INSTALL be the installation path of PetIBM. All header files for public PetIBM API are under $PETIBM_INSTALL/include/petibm, while those for third-party dependencies, if any, are under $PETIBM_INSTALL/include. For example, if users configure PetIBM using the flag --enable-amgxwrapper, the header file AmgXSolver.hpp will be copied to $PETIBM_INSTALL/include.

There are many headers of PetIBM, while only the top-level ones are normally used in an application code:

  • type.h: useful data types in PetIBM,
  • parser.h: functions to parse YAML input file,
  • mesh.h: abstract mesh object and its factory functions,
  • boundary.h: abstract boundary (and ghost points) object and its factory functions,
  • solution.h: abstract solution object and its factory functions,
  • operators.h: factory functions for creating operators (matrices),
  • linsolver.h: abstract object for linear solvers and its factory functions,
  • bodypack.h: abstract object for multi-body holders and its factory functions,
  • timeintegration.h: abstract object for temporal integration and its factory functions.

To compile with PetIBM headers, remember to have -I$PETIBM_INSTALL/include in the compilation flags. Also, the path to PETSc headers is necessary, e.g., -I$PETSC_DIR/PETSC_ARCH/include.

If users provide their own installations of third-party dependencies to build PetIBM, remember to also include those paths. For example, if users provide --with-yamlcpp-dir=$YAMLCPP_DIR during configuration, then they will have to add -I$YAMLCPP_DIR/include to the compilation flags.

Library

PetIBM provides two libraries, libpetibm.so and libpetibmapps.so (or their corresponding static libraries). libpetibm.so provides basic components for writing a solver without inheriting from any existing solvers. The mentioned liddrivencavity2d is an example of only using libpetibm.so. On the other hand, libpetibmapps.so provides basic flow solvers for derived classes. For example, the oscillatingcylinder case in api_examples uses libpetibmapps.so. Both libraries are located in $PETIBM_INSTALL/lib; use -L$PETIBM_INSTALL/lib -lpetibm -lpetibmapps to link.

There may be other libraries in $PETIBM_INSTALL/lib, depending on how users configure PetIBM. For example, if a user lets PetIBM installs yaml-cpp automatically, then the yaml-cpp library will be installed along with the PetIBM library in $PETIBM_INSTALL/lib. (Note: on some systems or OS, third-party libraries may be installed in $PETIBM_INSTALL/lib64.)

If users use CMake to build applications, using PetIBM libraries should be easy. PetIBM ships its own CMake config file. It means in CMake, just use find_package to find PetIBM and all dependencies (AmgX, PETSc, etc., except for MPI). For example (a snippet from the liddrivencavity2d API example):

set(CMAKE_PREFIX_PATH ${PETIBM_DIR} ${CMAKE_PREFIX_PATH})
find_package(MPI REQUIRED)
find_package(petibm 0.5.4 REQUIRED)
add_executable(liddrivencavity2d main.cpp)
target_link_libraries(liddrivencavity2d PRIVATE MPI::MPI_CXX petibm::petibm)

With the imported targets (e.g., MPI::MPI_CXX and petibm::petibm) in modern CMake, users usually don't have to specify include directories, link directories, and secondary dependencies as long as the upstream (i.e., the developers of MPI and PetIBM in this example) configures libraries correctly.

If manually linking PETSc library is desired, it can be found using pkg-config. For example, if, in a CMakeLists.txt, assuming PETSC_DIR refers to the PETSc installation, then do

set(CMAKE_PREFIX_PATH ${PETSC_DIR} ${CMAKE_PREFIX_PATH})
find_package(PkgConfig REQUIRED)
pkg_search_module(PETSC REQUIRED QUIET IMPORTED_TARGET petsc>=3.16)

Once pkg-config found PETSc, legacy CMake variables prefixed with PETSC_ will be available. Or, for modern CMake, an imported target PkgConfig::PETSC is available, too.

Alternatively, flags for linking against PETSc can be found in the file $PETSC_DIR/$PETSC_ARCH/lib/petsc/conf/petscvariables. (Look for the line starting with PETSC_WITH_EXTERNAL_LIB = .)

See the api_examples folder for more demo CMakeLists.txt files.

Compilers

If PetIBM was installed through conda/mamba, newer compilers may be needed to compile your own applications. For example, the Anaconda package of petibm=0.5.4=cuda114_<other tags> was built using gxx_linux-64=11.2 from conda-forge channel. That means when you build/compile your code against this specific PetIBM version, you may need to use g++ greater or equal to 11.2 to use newer glibc and glibcxx. The easiest way is to also use the latest tools from Anaconda, i,e., you may do (replace mamba with conda if you use the former one)

$ mamba install -c conda-forge "gxx_linux-64>=11.2" cmake make