azimonti / ma-libs

A set of libraries and tool in C++ and Fortran, developed during the years, that can be reused in other projects

Useful C++ / Fortran Libraries

This repository contains several libraries that I have developed through the years and that can be reused for other project.

Since the installation is not easy (in particular on Windows it requires both installation of Fortran and compilation of HDF5) I decided to create a dedicated repository, so that I can just link the relevant pre-compiled binaries whenever I need them.

C++

• C++ BLAS/LAPACK bindings made with templates supporting different types (float / double / std::complex<float> / std::complex<double> (built with --cmake-params "-DCPP_LIBALG_LAPACK=ON" or --build-suite).

• C++ neural network library with BLAS/LAPACK backend and HDF5 for storing the weights so that can be later reused (built with --cmake-params "-DCPP_LIBNN=ON" or --build-suite). Since it require BLAS/LAPACK will build CPP_LIBALG_LAPACK if not selected. Optionally Python bindings can be created with the optional arguement PYTHON_BINDINGS=ON (so --cmake-params "-DCPP_LIBNN=ON -DPYTHON_BINDING=ON"). A Python virtual environment can be created if needed with the script cpp/libnn/create_virtualenv.sh.

• C++ Python bindings allowing Python to call C++ (built with --cmake-params "-DCPP_PYTHON_BINDINGS=ON"). A detailed README is available here.

• C++ Fortran bindings allowing C++ to call Fortran (built with --cmake-params "-DCPP_FORTRAN_BINDINGS=ON").

• OpenGL multiplatform graphic engine, useful to visualize, useful fo data interpretation and presentation of numerical results in 2D and 3D (built with --cmake-params "-DCPP_LIBGRAPHIC_ENGINE=ON" or --build-suite).

Fortran

• HDF5 bindings library to hide the system call to HDF5 and write directly the requested type (build with -DF_LIBHDF5UTILS=ON or --build-suite).

• Fortran C++ bindings allowing Fortran to call C++ (built with --cmake-params "-DF_CPP_BINDINGS=ON").

• Linear algebra routines (build with -DF_LIBALGEBRA=ON or --build-suite).

• Algorithms routines (build with -DF_LIBALGO=ON or --build-suite).

• Testing library which produce output similar to doctest (build with -DF_LIBTEST=ON or --build-suite).

Required Tools

• Git
• CMake
• clang
• Fortran
• OpenGL

Quick Start

2) Download the external dependencies

./manage_dependency_libraries.sh --download

3) Compile the external dependencies

./manage_dependency_libraries.sh --build

They will be installed under ./usr/

If the libraries are used in several projects, it is recommended to define in .bash_profile or .bash_rc the following variables:

# export local libraries folders

export MA_LIBS_COMPILE_ROOT="MYPATH"

export USR_BUILD_DIR="${MA_LIBS_COMPILE_ROOT}/build"
export USR_LIB_DIR="${MA_LIBS_COMPILE_ROOT}/usr"
export USR_SRC_DIR="${MA_LIBS_COMPILE_ROOT}/src"

The compile with the flag --global:

./manage_dependency_libraries.sh --build --global

All projects will use this location (FindLIBS.cmake is looking for a global directory if a local is not present) and there is no need to recompile the dependencies at each installation of ma-libs.

3) Compile the libraries

The below will compile most of the libraries, specific instructions (e.g. for Python bindings) require the specific compilation flags described above.

./cbuild.sh --build-type Debug --build-suite
./cbuild.sh --build-type Release --build-suite

Executables can now be found under build/.

IMPORTANT NOTICE for WINDOWS

To properly link Intel Fortran libraries, ensure that the path to the Intel Fortran library directory is set in your environment. This path is necessary to locate ifconsol.lib, which is required by Intel Fortran for linking when building DLLs. Set the environment variable INTEL_FORTRAN_LIB_PATH to the directory containing ifconsol.lib, typically located at INTEL_INSTALL_PATH/Intel_fortran/compiler/latest/windows/compiler/lib/intel64_win. Failure to set this path may result in unresolved symbols during the build process.