Skip to content

Latest commit

 

History

History
103 lines (70 loc) · 3.31 KB

requirements.md

File metadata and controls

103 lines (70 loc) · 3.31 KB

Requirements

libffi

Linux

On Linux, this can be installed with your favorite package manager. For example, it can be installed on Ubuntu with:

apt-get install libffi-dev

macOS and OS X

On new versions of macOS, this comes preinstalled. If you find this to not be the case for you (this could potentially be an issue on OS X), it can be installed with Homebrew using:

brew install libffi

BLAS and LAPACK

Linux

Package Manager

On Linux, you can install BLAS and LAPACK with your favorite package manager. All that's expected is that you have libblas.so and liblapack.so.

For example, on Ubuntu this can be done with:

apt-get install liblapack-dev libblas-dev libgfortran3

You may need libgfortran if you built Fortran libraries. You can install that with:

apt-get install libgfortran3

MKL

Intel's Math Kernel Library, or MKL, contains math routines, including BLAS and LAPACK, that are specifically optimized for Intel processors. If the versions of BLAS and LAPACK mentioned above do not suit your needs, you may consider installing MKL. To install MKL, download the package from the Intel website and follow the instructions within the Install_Guide.pdf file. The particular library of interest that will be installed is libmkl_rt.so.

It is also important to setup the proper environmental variables, especially the LD_LIBRARY_PATH that specifies where to look for libmkl_rt.so; directions can be found here.

In order to use MKL in MAGICL, add :magicl.use-mkl to your *features* before compilation.

macOS and OS X

Homebrew

On macOS, Homebrew is the easiest way to get BLAS and LAPACK. You can install them with the following:

brew install gcc lapack

If you need gfortran, you can get that with

brew install gcc

By default, libgfortran is searched for where Homebrew installs it, namely /usr/local/opt/gcc/lib/gcc/7/. Likewise with libblas and liblapack, located in /usr/local/opt/lapack/lib/. Refer to CFFI's documentation on search paths to configure this more specifically. Libraries are searched here by default.

Accelerate

Another way to get BLAS and LAPACK on macOS is through the Accelerate framework. The required libraries are libBLAS.dylib and libLAPACK.dylib. While the Accelerate framework includes these libraries, they are not complete. We count more than 500 missing functions provided in the standard LAPACK distribution. In order to use the system-provided libraries, add :magicl.use-accelerate to your *features* before compilation.

MKL

Intel's Math Kernel Library is also available for macOS, and directions for installing it can be found here here.

In order to use MKL in MAGICL, add :magicl.use-mkl to your *features* before compilation.