Difference between revisions of "Matlab"

From ArchWiki
Jump to: navigation, search
(General installation)
(remove outdated information about non-64-bit versions, these are not sold anymore)
Line 7: Line 7:
{{Article summary wiki|Mathematica}}
{{Article summary wiki|Mathematica}}
{{Article summary end}}
{{Article summary end}}
{{Out of date|The Java-related instructions are seriously out of date: {{Pkg|openjdk6}} does not exist, {{AUR|bin32-openjdk6}} is outdated, etc.}}
From the [http://www.mathworks.com/products/matlab/ official website]:
From the [http://www.mathworks.com/products/matlab/ official website]:
Line 62: Line 60:
More details see [https://help.ubuntu.com/community/MATLAB MATLAB - Community Ubuntu Documents]
More details see [https://help.ubuntu.com/community/MATLAB MATLAB - Community Ubuntu Documents]
=== Install 32-bit Matlab on 64-bit system ===
{{Note|Follow this section only if you have a student version before R2013a. All releases since R2012b are 64-bit only.}}
These instructions are tested for the R2012a student version. To start off we need to install a bunch of packages from the [[multilib]] repository (this list may be out of date):
{{Pkg|lib32-mesa}}, {{Pkg|lib32-glu}}, {{Pkg|lib32-glibc}}, {{Pkg|lib32-libxmu}}, {{Pkg|lib32-zlib}}, {{Pkg|lib32-ncurses}}, {{Pkg|lib32-libxtst}}, {{Pkg|lib32-libxi}}, {{Pkg|lib32-libxrender}}, {{Pkg|lib32-libxfixes}}, {{Pkg|lib32-freetype2}}, {{Pkg|lib32-fontconfig}}, {{Pkg|lib32-libxdamage}}.
A few packages from the [[AUR]] might also be needed:
{{AUR|lib32-libxpm}}, {{AUR|lib32-libxp}}, {{AUR|bin32-openjdk6}} ({{Pkg|openjdk7}} might also work).
Once they are installed:
$ ./install -glnx86
After fixing any library dependency problems, you will be able to guide the installer through the install process.  If activation doesn't work in the installer you'll need to login to Wolfram's site and download a license file which you'll put in {{ic|${MATLAB}/licenses}}.  After the install process, you'll need to create a symbolic link because the installer oddly thinks it's installing the 64-bit version of Matlab:
$ cd ${MATLAB}/sys/java/jre
$ ln -s glnx86 glnxa64
In the above, ${MATLAB} should be replaced by the root directory you chose for your installation.
You'll also need to put another link in Matlab's {{ic|bin}} folder because despite asking Matlab to install the 32bit version, it still thinks you want to use the 64bit one:
$ cd ${MATLAB}/bin
$ ln -s glnx86 glnxa64
After that you'll need to edit the {{ic|${MATLAB}/bin/matlab}} script to get the java settings correct.  After the comments at the top of the file, you'll need the line:
export MATLAB_JAVA=/path/to/jre
Where {{ic|/path/to/jre}} will be {{ic|/opt/java/jre}} if you're using Sun's JRE.  The path is one level up from where {{ic|rt.jar}} is located.  That is, {{ic|rt.jar}} should be in {{ic|/opt/java/jre/libs}} if you used {{ic|/opt/java/jre}} for {{ic|MATLAB_JAVA}}. In Matlab R2010a it works pretty well, if you use the internal JRE with:
export MATLAB_JAVA=${MATLAB}/sys/java/jre/glnx86/jre
If using {{AUR|bin32-openjdk6}}, use
export MATLAB_JAVA=/usr/lib/jvm/java-7-openjdk/jre/
To run it, you'll need to:
$ ${MATLAB}/bin/matlab -glnx86 -desktop
At this point you may have to resolve more library dependencies or troubleshoot [[Java]].  Your version of Matlab may need a specific version of Sun's JRE installed.
=== Install supported compiler ===
=== Install supported compiler ===

Revision as of 10:46, 9 December 2013

Template:Article summary start Template:Article summary text Template:Article summary heading Template:Article summary wiki Template:Article summary wiki Template:Article summary wiki Template:Article summary end

From the official website:

MATLAB is a high-level language and interactive environment for numerical computation, visualization, and programming. Using MATLAB, you can analyze data, develop algorithms, and create models and applications. The language, tools, and built-in math functions enable you to explore multiple approaches and reach a solution faster than with spreadsheets or traditional programming languages, such as C/C++ or Java.

General installation

Note: Versions earlier than MATLAB R2013a are not compatible with Java 7. When installing an older version, first install libxp and jre6AUR.

MATLAB 2013a+ should install fine with the install script provided by MATLAB, without any additional packages.

To install from an iso file:

# modprobe loop
# mount -o loop matlab.iso /mnt/
# /mnt/install

Additionally, install xorg-fonts-100dpi, xorg-fonts-75dpi, and xorg-fonts-type1 so fonts are displayed properly in figures.

Create Desktop and Menu Shortcuts

If you choose Custom in Installation Type, the symlink will be created automatically. Otherwise, you may need to manually create a symlink in /usr/local/bin to make it easier to launch in terminal:

# ln -s /{MATLAB}/bin/matlab /usr/local/bin

To create a menu item, we need to get a icon first:

# wget http://upload.wikimedia.org/wikipedia/commons/2/21/Matlab_Logo.png -O /usr/share/icons/matlab.png

Then create a new .desktop file in /usr/share/applications with following lines:

#!/usr/bin/env xdg-open
[Desktop Entry]
Comment=Start MATLAB - The Language of Technical Computing
Exec=env -u _JAVA_OPTIONS matlab -desktop -nosplash -r "cd('%d'); edit '%f'"

The Exec command line is composed as follows:

  • -desktop is a flag needed to run matlab without a terminal.
  • -nosplash is a flag preventing the splash screen from showing and taking up a temporary space in your task bar.
  • env -u _JAVA_OPTIONS prevents the creation of a .java.log-file in your home directory by unsetting $_JAVA_OPTIONS. Java prints all options set this way to stdout, and matlab treats everything in stdout as error worth a log.
  • -r '...' executes matlab code on startup:
    1. cd('%d') This changes the working directory of matlab to the opened file’s directory.
    2. edit '%f' This spawns an editor tab for the file.

You can also put this .desktop file in the ~/Desktop directory to create a shortcut.

More details see MATLAB - Community Ubuntu Documents

Install supported compiler

In order for Matlab to work with C code (needed for simulink) it is necessarry to install a supported compiler. Install gcc44AUR from the AUR.

Then edit ${MATLAB}/bin/mexopts.sh and replace all occurances of CC='gcc' with CC='gcc-4.4' and CXX='g++' with CXX='g++-4.4'. Afterwards run

mex -setup

in Matlab and select the mexopts.sh file.


As one installs Matlab, it might complain that it can't find a package, for the most part just look at the package name and then install it with Pacman, or in the case of x86_64 there are some libraries only in AUR.

License: invalid machine id

The installer may complain about an invalid machine id, because it is looking for a network interface named eth0 to get a MAC address for activation, while new Arch Linux setups do not have a network interface called eth0 (systemd uses Predictable Network Interface Names).

To fix this change the name of the interface.

The machine id should now be different than 000000000000 and you should be able to install and activate MATLAB without problems.

Resolving start warnings/errors

  • Even if all needed libraries are installed, Matlab when starting can still report some missing libraries. This is resolved by symbolic linking of needed libraries to directories that Matlab checks at start-up. For example, if Matlab triggers error/warning about missing /lib64/libc.so.6 library, this can be resolved by:
# ln -s /lib/libc.so.6 /lib64
  • Matlab R2011b with an up-to-date Arch Linux (as of March 12, 2012) fails on startup with the familiar "Failure loading desktop class." A solution is to point Matlab to the system JVM (confirmed to work with the jdk7-openjdk package):
export MATLAB_JAVA=/usr/lib/jvm/java-7-openjdk/jre

MATLAB crashes when displaying graphics

To identify this error, start MATLAB with

LIBGL_DEBUG=verbose matlab

from the terminal and try to collect OpenGL information with opengl info from the MATLAB command prompt. If it crashes again and there is an output line like

libGL error: dlopen /usr/lib/xorg/modules/dri/swrast_dri.so failed 
version `GLIBCXX_3.4.15' not found (required by /usr/lib/xorg/modules/dri/swrast_dri.so))

then the problem is that MATLAB uses its own GNU C++ library, which is an older version than the up-to-date version on your Archlinux system. Make MATLAB use the current C++ library for your system by

cd /usr/local/MATLAB/R(your release)/sys/os/glnxa64
sudo unlink libstdc++.so.6
sudo ln -s /usr/lib/libstdc++.so.6

Blank/grey UI when using DWM/Awesome

wmname LG3D

OpenGL Acceleration

Once Matlab is installed type the following in the interpreter:

>> opengl info

The output should be similar to the following:

Version         = 2.0 Mesa 7.8.2
Vendor          = Advanced Micro Devices, Inc.
Renderer        = Mesa DRI R600 (RV710 9552) 20090101 x86/MMX/SSE2 TCL DRI2
MaxTextureSize  = 4096
Visual          = 0xcb (TrueColor, depth 24, RGB mask 0xff0000 0xff00 0x00ff)
Software        = false
# of Extensions = 107

Driver Bug Workarounds:
OpenGLBitmapZbufferBug    = 0
OpenGLWobbleTesselatorBug = 0
OpenGLLineSmoothingBug    = 0
OpenGLClippedImageBug     = 0
OpenGLEraseModeBug        = 0

This will vary with card you are using, but the important part is "Software = false". If it's not "false" for you, then there is a problem with your hardware acceleration.


With an NVIDIA card, the output should be similar to this:

>> opengl info
Version         = 2.1.2 NVIDIA 177.82
Vendor          = NVIDIA Corporation
Renderer        = Quadro NVS 140M/PCI/SSE2
MaxTextureSize  = 8192
Visual          = 0x21 (TrueColor, depth 16, RGB mask 0xf800 0x07e0 0x001f)
Software        = false
# of Extensions = 144

Driver Bug Workarounds:
OpenGLBitmapZbufferBug    = 0
OpenGLWobbleTesselatorBug = 0
OpenGLLineSmoothingBug    = 0
OpenGLClippedImageBug     = 1
OpenGLEraseModeBug        = 0

With the NVIDIA card all it required to start working was to copy the libGLU.so from the Matlab installed library ${MATLAB}/sys/opengl/lib/glnx86/libGLU.so to /usr/lib directory. Note that in x86_64 this might be a different directory.

Note: Although, not thoroughly tested, using lib32-nvidia-utils seems to work as well.