Compiling MRPT
This page describes how to
build MRPT from sources. If you want instead to install the binaries and getting quickly into developing your MRPT-based applications, go to the
download page and grab the
binaries for your system.
Please, report
errors or doubts on MRPT compilation to the
mrpt-users forum.
1. Prerequisites
1.1. On Windows
1.1.1 CMake
To install the CMake build system, follow the instructions at http://www.cmake.org/cmake/resources/software.html
In addition to the following notes, read the project's wiki: http://wiki.wxwidgets.org/MSVC
Also, if you want to build for 64bit (x64), be aware of the limited support of wxWidgets for this platform in Windows yet (read
this).
The source package for the wxWidgets library (named "wxALL" or "wxMSW", currently version 2.8.11) must be downloaded from
here and decompressed in a directory where CMake can find it (e.g. C:\wxWidgets).
In Windows
you'll have to manually modify the flag to compile the OpenGL support of wxWidgets. This is achieved by setting
wxUSE_GLCANVAS to 1 (the default is 0). This is the part of the file
wxwidgets\include\wx\msw\setup.h
that has to be modified:
// Setting wxUSE_GLCANVAS to 1 enables OpenGL support. You need to have OpenGL
// headers and libraries to be able to compile the library with wxUSE_GLCANVAS
// set to 1. Note that for some compilers (notably Microsoft Visual C++) you
// will need to manually add opengl32.lib and glu32.lib to the list of
// libraries linked with your program if you use OpenGL.
//
// Default is 0.
//
// Recommended setting: 1 if you intend to use OpenGL, 0 otherwise
#define wxUSE_GLCANVAS 1
Now, depending on your version of wxWidgets, open the Visual Studio project (which may be
wxwidgets\build\msw\wx_dll.dsw or
wxwidgets\build\msw\wx_vcX.sln). I recommend to build, with "Build all", only these configurations:
- DLL Unicode Release (or just "DLL Release" in wxWidgets >=2.9)
- DLL Unicode Debug (or just "DLL Debug" in wxWidgets >=2.9)
1.1.3 OpenCV
It is recommended to compile MRPT against
OpenCV 2.0.0 (or newer) in order to enjoy the new features such as SURF descriptors. See the instructions in: http://opencv.willowgarage.com/wiki/InstallGuide
One interesting option is to compile a SVN snapshot of OpenCV:
Checkout a fresh copy of the development branch by executing: (NOV-2009: OpenCV hosting changed, this is the new SVN uri:)
svn co https://code.ros.org/svn/opencv/trunk/opencv/ opencv
or use the graphical program TortoiseSVN.
Then follow the
compiling instructions.
Once compiled with CMake, running "Configure" in MRPT's CMake project will automatically detect the OpenCV build directory (where you have generated the projects with CMake) and will use it. If it's not detected, set
OpenCV_DIR to the build directory of OpenCV.
1.1.4 FFmpeg for Win32
The FFmpeg libraries are
optional, and they are only needed if you want to use
CFFMPEG_InputStream, mainly for supporting IP cameras.
Directly download and decompress anywhere in your disk :
- http://ffmpeg.arrozcru.org/builds/shared/ffmpeg-r16537-gpl-lshared-win32.tar.bz2
- or this local copy: http://babel.isa.uma.es/mrpt/downloads/ffmpeg-r16537-gpl-lshared-win32.tar.bz2
- or see the "Unofficial win32 builds" web for more recent releases: http://ffmpeg.arrozcru.org/autobuilds/
Then when running CMake (cmake-gui), enable
MRPT_HAS_FFMPEG_WIN32
, press "Configure" and then set
FFMPEG_WIN32_ROOT_DIR
to the directory where FFmpeg binaries have been uncompressed (e.g. "c:\ffmpeg-r16537-gpl-lshared-win32").
The FFmpeg DLLs will be required at runtime by programs compiled with MRPT under Windows, so make sure the directory
/bin is in the system PATH.
1.2. On GNU/Linux
1.2.1. Dependencies
These are the recommended libraries (the development packages!) that should exist in your system (see below for short instructions for Ubuntu/Debian/Fedora):
- OpenCV
- wxWidgets
- ffmpeg libraries (libavcodec, libswscale, etc.)
- Freeglut or glut
- zlib
- libftdi (optional)
- libdc1394-22
1.2.2. About OpenCV
You can compile MRPT against OpenCV 1.0.0, 1.1.0 or >=2.0.0, but it won't compile with OpenCV 1.1.0-pre1.
Install opencv libraries from your repositories or build it manually from SVN HEAD (I prefer this way...), by checking out a fresh copy of the development branch by executing:
svn co https://code.ros.org/svn/opencv/trunk/opencv/ opencv
And follow the compiling instructions.
Under Linux, you'll NOT need to do a "make install" after building OpenCV, since setting the OpenCV_DIR variable in CMake while compiling MRPT is enough.
1.2.2 Ubuntu, Debian
Invoke:
$ sudo apt-get install build-essential cmake libwxgtk2.8-dev \
libwxbase2.8-dbg libwxgtk2.8-dbg libftdi-dev libglut3-dev \
libhighgui-dev
(Note: You'll need to add libcvaux-dev, only if that package exists in your distribution and you haven't compile OpenCV manually)
If you have a modern Debian/Ubuntu, install this in addition to the packages above (they are not mandatory, but recommended):
$ sudo apt-get install libdc1394-22-dev libavformat-dev libswscale-dev
Note: If the version 2.8 (or newer) of wxWidgets is not available in your distribution, it must be installed manually.
1.2.3 Fedora
Execute:
$ su -c 'yum install gcc gcc-c++ make cmake wxGTK-devel opencv-devel freeglut-devel'
1.2.4 OpenSUSE
For OpenSUSE 10.X, invoke:
$ sudo zypper install make gcc gcc-c++ cmake cmake-gui pkg-config \
zlib-devel wxGTK-devel wxGTK-gl libusb-devel freeglut-devel
For OpenSUSE 11.X, invoke:
$ sudo zypper install make gcc gcc-c++ cmake cmake-gui pkg-config \
zlib-devel wxGTK-devel wxGTK-gl libusb-devel freeglut-devel
1.3. On Mac
Building under MacOS is not yet working 100%. Instructions will be given when we solve all the problems.
2. CMake build options
2.1. Using cmake-gui
Open cmake-gui (Available for Windows/Linux) and set the "source dir" to the root directory of the MRPT source package you have downloaded.
Set the "binary directory" to a new, empty directory where to generate the project files.
Press "configure", check for errors, tune the options as required (read below for a description of some options) and finally click "Generate".
2.2. Using cmake from the console
This choice is available for Windows/Linux/MacOS.
Go to a new, empty directory where to generate the project/Makefiles files and run:
$ cmake /home/.../MRPT
Replace "/home/.../MRPT" with your actual path to the MRPT source package you have just downloaded.
This will use all the default options. To tune them, invoke (available for Linux/MacOS):
$ ccmake .
(Click the image to see it at full size)
2.3. Interesting build options
For all platforms/compilers:
BUILD_SHARED_LIBS
: Build static libraries if set to OFF, or dynamic libraries (.so/.dll) otherwise. Default is ON, and it's strongly recommended to always use shared libs unless you have special need for static ones.
BUILD_EXAMPLES
: Whether you want to compile all the examples in the "/samples" directory. Default is OFF.
MRPT_ENABLE_EMBEDDED_ENABLED_PROFILER
: If enabled, all code blocks within macros "MRPT_BEGIN/MRPT_END" will be profiled and the statistics dumped to the console at the end of execution of any program. Default is OFF.
MRPT_ALWAYS_CHECKS_DEBUG
: If set to ON, additional security checks will be performed at run-time in many classes. Default is OFF.
MRPT_ALWAYS_CHECKS_DEBUG_MATRICES
: If set to ON, additional security checks will be performed at run-time in several Matrix operations. Default is ON.
MRPT_OCCUPANCY_GRID_CELLSIZE
: Can be either 8 or 16 (bits). The size of each cell in the class mrpt::slam::COccupancyGridMap2D. Default is 8 bits. More on this here.
USER_EXTRA_CPP_FLAGS
: You can add here whatever additional flags to be passed to the compiler.
MRPT_HAS_ASIAN_FONTS
: Enables Asian fonts in CMRPTCanvas (see wiki page), but increases library size by 1.5Mb. Default is OFF.
BUILD_xSENS
: Whether to use the CMT library for interfacing xSens inertial sensors. Default is ON.
For Windows only:
MRPT_HAS_FFMPEG_WIN32
: Enable this and (after running "Configure") then set FFMPEG_WIN32_ROOT_DIR to the directory where FFmpeg binaries have been uncompressed (e.g. "c:\ffmpeg-r16537-gpl-lshared-win32").
MRPT_HAS_BUMBLEBEE
: To enable integration of the Bumblebee stereo camera SDK. You'll need the vendor provided "Triclops" and "Digiclops" libraries. After setting this option to "ON", the new configuration fields "BUMBLEBEE_DIGICLOPS_ROOT_DIR" and "BUMBLEBEE_TRICLOPS_ROOT_DIR" will appear where the correct corresponding paths must be entered.
For GNU GCC compiler only:
MRPT_ENABLE_LIBSTD_PARALLEL_MODE
: Enables the GNU libstdc++ parallel mode (See http://gcc.gnu.org/onlinedocs/libstdc++/manual/parallel_mode.html). Default is OFF.
MRPT_ENABLE_PROFILING
: Enables generation of information required for profiling. Default is OFF.
MRPT_OPTIMIZE_NATIVE
: Enables optimization for the current architecture (-mtune=native). Default is OFF for old GCC versions, ON for 4.2+. If you have an old version of GCC (<4.2), this option cannot be set since it's not recognized by the compiler. Instead, set USER_EXTRA_CPP_FLAGS to the optimization flags for your platform, for example: -march=pentium4.
3. Generate Makefiles/IDE Projects
Just select your preferred Makefile/IDE system with CMake. Supported targets are 32/64 bit Visual Studio projects, Unix Makefiles, Codeblocks projects with MinGW, MinGW Makefiles, etc...
4. Compile
Just build as usual: from Visual Studio click on "Build all" for the targets you want to build (Debug or Release), for Unix Makefiles invoke:
$ make -j2
After building everything, it's a good idea to run the tests by building the "test" target under Visual Studio or by "make test" in Unix/MacOS.
5. Special instructions for MinGW
- Install MinGW: Recommended: http://www.tdragon.net/recentgcc/
- Before compiling MRPT with MinGW, it's strongly recommended to compile wxWidgets and OpenCV with MinGW:
- Build wxWidgets. Open a command prompt and go to the directory build/msw. Then execute the following commands to rebuild the Release and Debug configurations (as shared libs), so CMake can correctly detect wxWidgets:
mingw32-make -f makefile.gcc SHARED=1 USE_OPENGL=1 BUILD=release
mingw32-make -f makefile.gcc SHARED=1 USE_OPENGL=1 BUILD=debug
- Build OpenCV. Use its CMake build system, select the MinGW compiler and follow the generic OpenCV compilation instructions.
- Open cmake-gui and select the source directory and an empty target (binary) directory.
- Press configure and in the compilers dialog pick MinGW Makefiles. If you obtain an error like:
CMake Error: CMake was unable to find a build program corresponding to "MinGW Makefiles".
CMAKE_MAKE_PROGRAM is not set. You probably need to select a different build tool.
it means MinGW is not correctly installed in the system. Review the installation process described above.
- If everything goes fine, you'll see the new CMake variables remarked in red. Go through the normal configuration process for MRPT, and when you are satisfied, press Generate.
- Open a console and in the newly created binary directory, invoke:
mingw32-make
either by writing the full path (e.g. c:\MinGW\bin\mingw32-make) or by adding the "bin" directory of your MinGW installation to the system PATH. This should start the normal build process.