##########################################
Install Graphics and Compute on Linux Mint
##########################################

OpenGL
------

To use OpenGL on Linux, the X11 Input extension library, the X11 miscellaneous
utility library, an OpenGL API implementation, and
the OpenGL Utility library are needed.

.. code:: bash

   sudo apt-get install libxi-dev libxmu-dev libgl1-mesa-dev libglu1-mesa-dev xorg-dev

The Mesa implementation by default creates a symbolic link to a nonexistent
:file:`libGL.so.1.2.0`; change that link to

.. code:: bash

   sudo ln -s -f /usr/lib/libGL.so.<#> /usr/lib/x86_64-linux-gnu/mesa/libGL.so

The installation is now complete, but the previous step may be needed each time
a proprietary driver (e.g. AMD, Nvidia, Intel) is installed because the new
driver may overwrite the symbolic link.

One can also install `GLEW`_ and `GLFW`_ to make OpenGL development easier.  The
former uses a Makefile while the latter uses CMake and Linux's
:program:`xorg-dev` library.  If CMake fails to locate OpenGL, run

.. code:: bash

   sudo apt-get install freeglut3-dev

.. _GLEW: http://glew.sourceforge.net/
.. _GLFW: http://www.glfw.org/

:doc:`Headless Rendering </blog/2013/01/11/set-up-system-programming-environment>`
----------------------------------------------------------------------------------

`EGL`_ is an alternative way to run OpenGL `headless`_.  Its header files can be
found in the registry, and the corresponding library implementation are
installed with the GPU drivers.  In this case, only the `OpenGL headers`_ are
needed.  :file:`gl.h`, :file:`glx.h`, and :file:`wgl.h` need to be extracted
from the GPU drivers e.g. *--opengl-headers* for Nvidia's installation script.
Within a virtualized environment, the libraries need to be installed manually.

.. code:: bash

   sudo apt-get install libegl1-mesa-dev libgles2-mesa-dev

.. _EGL: https://www.khronos.org/registry/EGL/
.. _headless: https://devblogs.nvidia.com/parallelforall/egl-eye-opengl-visualization-without-x-server/
.. _OpenGL headers: https://www.khronos.org/registry/OpenGL/index_gl.php

Vulkan
------

Do not attempt to use `OpenGL multithreading`_ and only use `multiple contexts`_
for data streaming.  The Vulkan drivers are vendor specific and are typically
installed with the GPU drivers.  The `Vulkan header files`_ are at the registry.
The `Vulkan validation layers`_ require the use of the `Vulkan SDK`_.

.. _Vulkan: https://www.khronos.org/vulkan/
.. _OpenGL multithreading: https://en.wikipedia.org/wiki/Vulkan_(API)
.. _multiple contexts: https://blog.gvnott.com/some-usefull-facts-about-multipul-opengl-contexts/
.. _Vulkan header files: https://www.khronos.org/registry/vulkan/#headers
.. _Vulkan validation layers: https://vulkan-tutorial.com/Drawing_a_triangle/Setup/Validation_layers#page_Using_validation_layers
.. _Vulkan SDK: https://www.lunarg.com/vulkan-sdk/

CUDA Toolkit
------------

This is no longer necessary if :doc:`nvidia-docker </blog/2016/09/11/tensorflow-tensorboard-and-docker>`
is used.  Otherwise, prefer the **.sh** Linux flavor of `CUDA toolkit`_.  The
most important step is to avoid installing the drivers that comes with the
package: the :doc:`latest proprietary Nvidia drivers </blog/2013/01/03/install-nvidia-drivers-on-linux-mint>`
should be used instead.

.. _CUDA toolkit: https://developer.nvidia.com/cuda-downloads

OpenCL
------

The CUDA toolkit also comes with an implementation of OpenCL.  The
`OpenCL headers`_ can be found at the Khronos registry.

.. _OpenCL headers: https://www.khronos.org/registry/OpenCL/