OpenVSIP uses a build system that is based on the GNU autotools. A typical build process thus consists of executing configure, make, and make install. The individual steps are discussed in detail below.
OpenVSIP can be compiled for a variety of target platforms, including x86, ppc, and cuda. It relies on (partly optional) third-party libraries for specific computational kernels:
and for additional language bindings:
- Python
- [NumPy] (http://www.numpy.org/)
- [Boost] (http://www.boost.org/)
OpenVSIP is known to compile with G++, version 4.8 and higher. It is being developed on Fedora.
Testing and benchmarking requires
To configure OpenVSIP:
% mkdir objdir
% cd objdir
% srcdir/configure [options]
Useful options:
--prefix=dirname: Specify the toplevel installation directory. (The default is/usr/local.)--enable-mpi: Enable support for the Parallel VSIPL++ API.--enable-lapack=<lapack>: Enable LAPACK bindings using the specified backend.--enable-fft=<fft-backend-list>: Enable the specified FFT backends.
To build OpenVSIP:
% make
To run the OpenVSIP test suite:
% make check
(To run tests in parallel, use make check parallelism=<n>
with the specified concurrency level.)
To install OpenVSIP:
% make install
This will install OpenVSIP in the prefix specified during configuration.
OpenVSIP contains a number of example applets that demonstrate how to use specific APIs. To build them, start by setting up an OpenVSIP workspace:
% $(prefix)/bin/ovxx-create-workspace <workspace>
Within the created workspace directory structure you can now run make
(or make <target>) to build the demo apps. Make sure to set your environment
variables (notably LD_LIBRARY_PATH) so that the appropriate libraries are found.
For an architectural overview of OpenVSIP please refer to Architecture