Parallel Build
This page only applies to SU2 version 6.2.0 and lower.The new documentation section can be found here.
SU2 uses the GNU automake tools to configure and build the software from source in Linux and Mac OS X environments (or with a terminal emulator on Windows). While a basic version of the code can easily be compiled, we recognize that many users will want to extend capabilities by running calculations on multiple cores and on distributed memory clusters. Luckily, the build process is largely the same, apart from a few new options and dependencies.
To illustrate an advanced build, let’s assume that you would like to build SU2 for running parallel calculations. In short, you will need to make sure some extra software is available and then execute a set of commands like the following:
$ cd /path/to/SU2
$ ./bootstrap
$ ./configure --prefix=/path/to/install/SU2 CXXFLAGS="-O3" --enable-mpi
--with-cc=/path/to/mpicc --with-cxx=/path/to/mpicxx
$ make -j 8 install
Let’s break this down and discuss the configure process and options in more detail.
Bootstrap
Before building, a bootstrap script is included in the SU2/ directory for quickly customizing the autotools structure on your system. Simply run
$ ./bootstrap
in the SU2/ root directory (you may need to adjust your PATH after running this script) to build the appropriate dependencies and reset the makefile structure rather than calling autoreconf on your own.
Compiler Flags
You can submit flags to your compiler for building SU2 using the CXXFLAGS
variable. The most common choices are to impose a level of compiler optimization or perhaps a debug flag with -g
. For example, a high level of compiler optimization can be set by adding
CXXFLAGS="-O3"
to the configure
call.
Parallel Support
First, to build in parallel, we need to inform the configure process by including the following options:
--enable-mpi --with-cc=/path/to/mpicc --with-cxx=/path/to/mpicxx
These three options enable parallel support and specify the MPI implementation that you would like to use for building. Note here that your machine must have an implementation of the MPI standard installed, i.e., mpicc
and mpicxx
must be installed on your system. For example, common MPI flavors used by the development team are Open MPI, MPICH, and Intel MPI. Additionally, mesh partitioning software is needed to decompose the meshes when running in parallel. In order to simplify the build process, the ParMETIS graph partitioning software ships with the SU2 source, and it will be compiled and linked automatically for you when the options above are prescribed.
The --prefix
option defines the location that the executables will be installed. They will be placed in a folder named bin/ within your chosen install location from –prefix. If the --prefix
option is not specified, the code will be installed in /usr/local/bin
, which may require admin access. Note that we are also using the “-j N” option of the make command in order to compile SU2 in parallel using N cores. This can greatly reduce the compilation time if building on a multicore laptop, workstation, or cluster head node. While not required, here we are combining the make
and install
commands into one.
Make sure to note the SU2_RUN and SU2_HOME environment variables displayed at the conclusion of configure. It is recommended that you add the SU2_RUN and SU2_HOME variables to your ~/.bashrc file and update your PATH variable to include the install location ($SU2_RUN, specified by --prefix
).
- Previous
- Next