Build SU2 on Linux/MacOS

For information on how to build older versions of SU2, have a look here.

Note that the following guide works only on Linux/MacOS and on Windows using Cygwin or the Linux Subsystem.

Quick Compilation Guide

This is a quick guide to compile and install a basic version of SU2. For more information on the requirements and a more detailed description of the build system continue reading the rest of this page.

Short summary of the minimal requirements:

  • C/C++ compiler
  • Python 3

Note: all other necessary build tools and dependencies are shipped with the source code or are downloaded automatically.

If you have these tools installed, you can create a configuration using the found in the root source code folder:

./ build

Use ninja to compile and install the code

./ninja -C build install



Installing SU2 from source requires a C++ compiler. The GNU compilers (gcc/g++) are open-source, widely used, and reliable for building SU2. The Intel compiler set has been optimized to run on Intel hardware and has also been used successfully by the development team to build the source code, though it is commercially licensed. The Apple LLVM compiler (Clang) is also commonly used by the developers.

  • GNU gcc / g++
  • Intel icc / icpc
  • Apple LLVM (clang)

Note: SU2 uses some C++11 features, that means at least GCC >= v4.7, Clang >= v3.0 or Intel C++ >= v12.0 is necessary.


In order to build SU2 with parallel support, you need a suitable MPI installation on your machine. During the configuration the build tool does a check (using pkg-config) and enables MPI support. If no installation is found, a serial version of SU2 will be compiled. It is possible to force the MPI mode with the meson option -Dcustom-mpi=true, it is then assumed that the compilers and/or the environment variables have the right flags, include directories, and linker arguments.

Note: Problems have been reported with MPICH where the build system does not detect the MPI installation, this can be solved using the meson options -Dcustom-mpi=true -Dextra-deps=mpich. If MPI was installed in a user directory, ensure also that the environment variable PKG_CONFIG_PATH is correctly setup.


SU2 requires Python 3 for compilation and for running the python scripts. Make sure that you have properly set up the python3 executables in your environment.

Optional: swig and mpi4py

If you want to use the python wrapper capabilities, also swig and mpi4py are required. On Linux swig should be available in the package manager of your distribution and mpi4py can be installed using pip.

On Mac OS X, you can use the Homebrew package manager. Once it is installed on your system, you can install Swig by running:

$ brew install swig

Install mpi4py with Python pip using easy install:

$ easy_install pip
$ pip install mpi4py

Automatically installed dependencies

The following dependencies are automatically downloaded (or initialized if source code was cloned using git) during the configuration.

Meson and Ninja

The build system of SU2 is based on a combination of meson (as the front-end) and ninja (as the back-end). Meson is an open source build system meant to be both extremely fast, and, even more importantly, as user friendly as possible. Ninja is a small low-level build system with a focus on speed.

CoDiPack and MeDiPack

In order to use the discrete adjoint solver the compilation requires two additional (header-only) libraries. CoDi provides the AD datatype and MeDi provides the infrastructure for the MPI communication when the reverse mode of AD is used.

Configuration and Compilation

Like mentioned above, SU2 uses meson and ninja for configuration and compilation, respectively. A configuration using meson is generated first and then an invocation of ninja is used to compile SU2 with this configuration.

Basic Configuration

In the root folder of the sources you will find a python script called This script generates a configuration. It will also check whether all dependencies are found and downloads some of them if necessary see previous section.

Note: For the following steps you can also use preinstalled versions of meson and ninja available on your machine. Just replace the ./ and ./ninja calls with the binaries of the respective installations. However, this way you have to manually make sure that the correct versions of CoDiPack and MeDiPack are placed in the externals/ folders.

The only required argument for is a name of a directory where it should store the configuration. You can have multiple configurations in different folders next to each other. To generate a basic configuration that will be stored in the folder build use

 ./ build

Options can be passed to the script to enable or disable different features of SU2. Below you find a list of project options and their default values:

Option Default value Description
-Denable-autodiff false enable AD (reverse) support (needed for discrete adjoint solver)
-Denable-directdiff false enable AD (forward) support
-Denable-pywrapper false enable Python wrapper support
-Dwith-mpi auto Set dependency mode for MPI (auto,enabled,disabled)
-Dwith-omp false enable MPI+Threads support (experimental)
-Denable-cgns true enable CGNS support
-Denable-tecio true enable TECIO support
-Denable-mkl false enable Intel MKL support
-Denable-openblas false enable OpenBLAS support
-Denable-pastix false enable PaStiX support
-Denable-mpp false enable Mutation++ support
-Denable-mixedprec false enable the use of single precision on linear solvers and preconditioners

For example to enable AD support pass the option to the script along with a value:

./ build -Denable-autodiff=true

To set a installation directory for the binaries and python scripts, use the --prefix option, e.g.:

./ build -Denable-autodiff=true --prefix=/home/username/SU2

If you are not interested in setting custom compiler flags and other options you can now go directly to the Compilation section, otherwise continue reading the next section.

Advanced Configuration

In general meson appends flags set with the environment variable CXXFLAGS. It is however recommended to use mesons built-in options to set debug mode, warning levels and optimizations. All options can be found here or by using ./ configure. An already created configuration can be modified by using the --reconfigure flag, e.g.:

./ build --reconfigure --buildtype=debug

Note that it is only possible to change one option at once.

Build Type

The debug mode can be enabled by using the --buildtype=debug option. This adds -g flag and disables all compiler optimizations. If you still want to have optimizations, use --buildtype=debugoptimized. The default build type is release.

Compiler optimizations

The optimization level can be set with --optimization=level, where level corresponds to a number between 0 (no optimization) and 3 (highest level of optimizations) which is the default. However, that may not result in optimum performance, for example with the GNU compilers level 2 and the extra flag -funroll-loops results in better performance for most problems.

Some numerical schemes support vectorization (see which ones in the Convective Schemes page), to make the most out of it the compiler needs to be informed of the target CPU architecture, so it knows what “kind of vectorization” it can generate (256 or 512bit, 128bit being the default). With gcc, clang, and icc this can be done via the -march=?? and -mtune=?? options, where ?? needs to be set appropriately e.g. skylake, znver3, etc., these flags can be passed to the compiler by setting CXXFLAGS before first running meson (which will print some messages acknowledging the flags).

Warning level

The warning level can be set with --warnlevel=level, where level corresponds to a number between 0 (no warnings) and 3 (highest level of warning output). Level 1 corresponds to -Wall, level 2 to -Wall -Wextra and level 3 to -Wall -Wextra -Wpedantic. The default level is 0.

Note: The warning flags -Wno-unused-parameter, -Wno-empty-body and -Wno-format-security are always added by default.

Linear algebra options

Compiling with support for a BLAS library (-Denable-mkl or -Denable-openblas) is highly recommended if you use the high order finite element solver, or radial basis function (RBF) interpolation in fluid structure interaction problems. Linear solvers and preconditioners can be accelerated with option -Denable-mixedprec=true, which will switch those computations to single precision while all other aspects of SU2 remain in double precision, for fluid simulations this does not reduce accuracy since the solution is iterative. However, large structural FEA problems may be adversely affected. To a lesser extent MKL 2019 is also used to accelerate (~5%) sparse linear algebra operations. -Denable-mkl takes precedence over -Denable-openblas, the system tries to find MKL via pkg-config, if that fails it will then look for MKL in /opt/intel/mkl, this can be changed via option -Dmkl_root. When OpenBLAS support is requested the build system uses pkg-config to search the system for package openblas, option -Dblas-name, if the library was built from source it may be necessary to set the environment variable PKG_CONFIG_PATH.

For large structural FEA problems on highly anisotropic grids iterative linear solvers might fail. Version 7 introduces experimental support for the direct sparse solver PaStiX (-Denable-pastix) see detailed instructions in TestCases/pastix_support/readme.txt.

If the use of BLAS is restricted to RBF interpolation, parallel versions of OpenBLAS can be used, the number of threads will then have to be controlled via the appropriate environment variable (consult the OpenBLAS documentation). Otherwise sequential BLAS should be used.

Note: The BLAS library needs to provide support for LAPACK functions. If this is not the case, the linker will fail with “undefined reference” errors, this problem can be solved by installing LAPACK and specifying it as an extra dependency when running using -Dextra-deps=lapack (this uses pkg-config, use commas to separate the names of multiple extra dependencies).


Finally to compile and install SU2 use

./ninja -C build install

where build is again a folder with a configuration created using a call to described in the previous section. By default ninja uses all available cores in your system for the compilation. You can set the number of cores manually by using the -jN flag, where N is the number of cores you want to use.

Setting environment variables

Set the environment variables to use the executables from any directory without explicity specifying the path as described in the installation section.


MPI installation is not found

Meson looks for an MPI installation using pkg-config. But if your MPI implementation does not provide them, it will search for the standard wrapper executables, mpic, mpicxx, mpic++. If these are not in your path, they can be specified by setting the standard environment variables MPICC, MPICXX during configuration.

mpi4py library is not found

Meson imports the mpi4py module and searches for the include path. If it is installed in a custom location, make sure to add this path to the PYTHONPATH environment variable prior calling

Ninja compiles but fails to install

If building on a cluster that uses a NFS filesystem, ninja may finish the compilation but fail to install with an error such as:

OSError: [Errno 22] Invalid argument: 'SU2_CFD/src/SU2_CFD'

This is a known bug in earlier versions of Python 3. Try upgrading to Python >= 3.7 then rerun ninja.

Improve this page