NLopt Installation

From AbInitio

(Difference between revisions)
Jump to: navigation, search
Revision as of 19:55, 13 November 2008 (edit)
Stevenj (Talk | contribs)
(Octave and Matlab plugins)
← Previous diff
Revision as of 19:57, 13 November 2008 (edit)
Stevenj (Talk | contribs)
(Changing the installation directory)
Next diff →
Line 29: Line 29:
cc -I$HOME/install/include myprogram.c -L$HOME/install/lib -lnlopt -lm -o myprogram cc -I$HOME/install/include myprogram.c -L$HOME/install/lib -lnlopt -lm -o myprogram
 +
 +See also below for how to change the installation directories for Octave and Matlab plugins, if you are installing those.
==Octave and Matlab plugins== ==Octave and Matlab plugins==

Revision as of 19:57, 13 November 2008

NLopt
Download
Release notes
FAQ
NLopt manual
Introduction
Installation
Tutorial
Reference
Algorithms
License and Copyright

The installation of NLopt is fairly standard and straightforward, at least on Unix-like systems (GNU/Linux is fine). It doesn't require any particular packages to be installed except for a C compiler, although you need to have w:GNU Octave and/or Matlab installed if you want to install the Octave and/or Matlab plugins, respectively.

In particular, NLopt uses the standard w:GNU Autoconf configure script, which means that you compile it via:

./configure
make

in the nlopt directory. Then, you would switch to be the roInsert code hereot user, or use the sudo command, to install the NLopt libraries and header files via:

make install

By default, this installs the NLopt static library (libnlopt.a) in /usr/local/lib and the NLopt header file (nlopt.h) in /usr/local/include, as well manual pages and a few other files.

In the following, we describe a few details of this installation process, including how to change the installation location.

Contents

Changing the installation directory

You may wish to install NLopt in a directory other than /usr/local, especially if you do not have administrator access to your machine. You can do this using the --prefix argument to the configure script.

For example, suppose that you want to install into the install subdirectory of your home directory ($HOME). You would do:

./configure --prefix=$HOME/install
make
make install

This will create the directories $HOME/install/lib etcetera and install NLopt into them. However, now when you compile code using NLopt, you will need to tell the compiler where to find the NLopt header files (using -I) and libraries (using -L) with something like:

cc -I$HOME/install/include myprogram.c -L$HOME/install/lib -lnlopt -lm -o myprogram

See also below for how to change the installation directories for Octave and Matlab plugins, if you are installing those.

Octave and Matlab plugins

When you compile NLopt using the above commands, it will automatically compile plugins for both Matlab and GNU Octave (a free Matlab clone) if the latter programs are installed.

Matlab

In particular, for Matlab plugins to be installed, you need to have the Matlab mex compiler command in your Unix PATH. Some versions of Matlab also require that you compile NLopt as a shared library in order to produce a Matlab plugin; see below.

By default, the Matlab plugins (along with help files and other .m files) are installed into the MATLABPATH printed out by matlab -n (which gives some directory within your Matlab installation directory), so that they will be available to all Matlab users. You can override this default (e.g. if you don't have administrator access on your machine) by passing a MEX_INSTALL_DIR to configure, via:

./configure MEX_INSTALL_DIR=dir

to install the Matlab plugins in directory dir. In this case, however, when you run Matlab you will either need to run in the dir directory or explicitly add dir to your Matlab path (see the Matlab path command).

Octave

For the Octave plugins to be installed, you need to have the Octave mkoctfile program in your PATH. mkoctfile is Octave's equivalent of mex. If you are using a GNU/Linux system, and you installed Octave using one of the precompiled packages for your distribution, then you probably need to install a separate package to get mkoctfile. For example, on Debian you need to install the octave-headers package, and on Redhat you need the octave-devel package.

By default, the compiled Octave plugins (.oct files) are installed into the systemwide site/oct directory (usually something like /usr/lib/octave/2.1.73/site/oct/i486-pc-linux-gnu), and the .m script files are installed into the systemwide site/m directory (usually something like /usr/share/octave/2.1.73/site/m/). You can change these defaults by passing OCT_INSTALL_DIR and M_INSTALL_DIR, respectively, to the configure script, via:

./configure OCT_INSTALL_DIR=octdir M_INSTALL_DIR=mdir

In this case, however you will either need to run Octave in the directory where these files are installed or explicitly add those directories to the Octave path (see the Octave path command).

Shared libraries

By default, NLopt compiles as a static library. This means that each program you link to NLopt will make a separate copy of the library, which wastes a little disk space. The alternative is to compile NLopt as a shared library (also called a dynamic-link library). While more efficient in terms of disk space etcetera, shared libraries require a bit more care to install properly, which is why we don't install them by default.

Compiling NLopt as a shared library is easy. Just add --enable-shared to the configure flags, as in:

./configure --enable-shared

Then you run make and make install as usual.

However, at this point you need to tell the operating system where to find the shared library, so that the runtime linker works properly. There are at least two ways to do this. First, you can use the LD_LIBRARY_PATH environment variable. For example, if you installed into the /foo/bar directory, so that the library is in /foo/bar/lib, then you would do

export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/foo/bar/lib

in the bash shell, or

setenv LD_LIBRARY_PATH ${LD_LIBRARY_PATH}:/foo/bar/lib

in csh or w:tcsh.

Alternatively, in GNU/Linux systems, you can add the library directory to the system-wide file /etc/ld.so.conf and then (as root) run /sbin/ldconfig.

NLopt with C++ algorithms

NLopt, as-is, is callable from C, C++, and Fortran, with optional Matlab and GNU Octave plugins. By default, it includes only subroutines written in C (or written in Fortran and converted to C), to simplify linking. If you configure with:

./configure --with-cxx

however, it will also include algorithms implemented in C++ (currently, just the StoGO algorithm), and the resulting library will be called libnlopt_cxx and is linked with -lnlopt_cxx.

The libnlopt_cxx has the same interface as the ordinary NLopt library, and can still be called from ordinary C and Fortran programs. However, to use it you must also link with the C++ standard libraries. The easiest way to do this is to link with the C++ linker: compile your source files into .o object files, and then call the C++ compiler to link these .o files with -lnlopt_cxx into your executable program.

It is because this linking process is somewhat annoying, and it only adds a single more algorithm (StoGO) to NLopt, that by default we omit StoGO to create a library that does not require the C++ standard libraries to link.

Personal tools