Skip to content

Latest commit

 

History

History
122 lines (84 loc) · 4.38 KB

install.md

File metadata and controls

122 lines (84 loc) · 4.38 KB
Raw

Installing MDI++

This program has various build-time requirements, these are:

  1. C++11 compiler, GCC or Clang have been tested under OSX and Ubuntu.

  2. GNU make, limited GNU features are used, but I haven't tested with anything else.

  3. pkg-config, used within config.mk to auto-detect the location of headers

  4. Boost, only the program_options module is used to parse command line parameters at the moment.

  5. Eigen 3, is used for all linear algebra within MDI++.

  6. optional: CUDA, is used for enable GPU based computation.

What you have to do to get these various packages installed on your computer will depend; I'll start by assuming that the reader is running Ubuntu.

Ubuntu

Installation under Ubuntu is easiest, after downloading the code, running the following commands should suffice to build MDI++:

$ sudo apt-get install build-essentials libboost-all-dev libeigen3-dev
$ sudo apt-get install nvidia-cuda-toolkit  # only needed if GPU support wanted

and all should be OK. If you want to use GPU acceleration, you will need the CUDA toolkit to be installed (I'm currently using 6.5). There are various ways of going about this, but for reference under OSX I used cuda_6.5.14_mac_64.pkg from NVIDIA and under my recent version of Ubuntu it is already included and can be installed with:

$ sudo apt-get install nvidia-cuda-toolkit

Some modification of config.mk may be needed to get it building with CUDA, there are some commented sections that work for my various configurations.

OS X

There are a variety of package managers available under OS X, common alternatives are Fink, MacPorts and Homebrew, but I personally use Homebrew so the following instructions will target this. Note that you should expect to encounter incompatibilities if you install more than one of the above systems.

Installing the required C++ libraries under Homebrew is similar to Ubuntu:

$ brew install --c++11 boost eigen

Additionally pkg-config is needed in order to locate the relevant header files:

$ brew install pkg-config

You will need a working compiler to install these packages, however as this is a requirement for homebrew to function this doesn't need to be installed explicitly.

Should you desire CUDA support, things are somewhat complicated due to unfortunate interactions with the clang derived compiler currently provided by Apple using an incompatible stdlib. The easiest way I have found of getting CUDA support working under OSX is by using GCC, you would therefore run:

$ brew install gcc-4.9 pkg-config
$ brew install --c++11 --cc=gcc-4.9 --build-from-source boost eigen

which will install the compiler and required pkg-config, then build and install the C++ libraries using the correct conventions.

Redhat Linux

Redhat follows a similar naming convention to Ubuntu. Boost and Eigen can be installed with the following commands:

# yum install boost boost-devel
# yum install eigen3 eigen3-devel

then you probably need to get an update compiler installed as the default included in RedHat is very out of date. This can be accomplished by running:

# yum install devtoolset-2-toolchain

devtoolset-3 is also available, and can be used if you are able to use it. Finally the build command is somewhat different, as you need to run it "within" Redhat's Developer Toolset:

$ scl enable devtoolset-2 'make ncuda=1'

We don't have an Nvidia card on this box so therefore can't comment on installing CUDA inside Redhat. If you have success please let me know and I'll update this document and other people can benefit.

Building MDI++

Once the relevant compilers and supporting libraries are installed, you should be able to build MDI by typing make. Should it fail, some build time configuration options are available in config.mk. By default it will try and compile a debug build with CUDA support enabled, but these can be turned off with various preprocessor macros. Should you not want to build with CUDA support, you can set the preprocessor macro ncuda, for example:

$ make ncuda=1

The makefiles also support opt for turning on optimisations, due to optimisations within Eigen it is strongly recommended to build with this options when analysing larger datasets:

$ make opt=1

the ncuda and opt flags can both be passed to turn off CUDA support while turning on compiler optimisations.