https://github.com/thequackdaddy/openblasr

Compiling openblas for R on Windows
https://github.com/thequackdaddy/openblasr

data-science mingw-w64 openblas r windows

Last synced: 18 days ago
JSON representation

Compiling openblas for R on Windows

• Host: GitHub
• URL: https://github.com/thequackdaddy/openblasr
• Owner: thequackdaddy
• License: bsd-3-clause
• Created: 2018-02-14T17:09:39.000Z (almost 7 years ago)
• Default Branch: master
• Last Pushed: 2019-02-17T05:00:12.000Z (almost 6 years ago)
• Last Synced: 2024-10-17T09:28:03.442Z (2 months ago)
• Topics: data-science, mingw-w64, openblas, r, windows
• Language: PowerShell
• Homepage:
• Size: 49.8 KB
• Stars: 4
• Watchers: 2
• Forks: 3
• Open Issues: 0
• Metadata Files:
  • Readme: README.rst
  • License: LICENSE

Awesome Lists containing this project

README

        
|Appveyor Build Status|

Overview

========

This project is used to compile openblas_ on Windows so that it can be used

as a faster BLAS alternative to the default BLAS provided with R_ on Windows.

This is used in the companion r-openblas_ tools--an openblas optimized 64-bit R

for Windows.

Acknowledgement

===============

First, I am grateful to Avraham Adler for his excellent `blog post`_ covering

this topic. If you want to read his script--which this package mostly follows--

please do so.

Procedure

=========

Prerequisites:

The key difficultly in getting this to work is R is built using the GNU tools

which have sometimes spotty and confusing windows support. However, the key

elemenets are that you need to download the following:

- Rtools_ Used for the actual ``gcc`` and ``gfortran`` compilers.

- MSYS2_ The compilation environment. MSYS2 ``bash`` must be used.

- Appveyor_ Is an online continuous integration system that essentially runs

Windows buils.

Strategy:

GCC can be configured either through a ``Makefile`` or using environment

variables. In this example, I'll use environment variables to avoid having

to copy files around. We can configured the environment variables using

an ``appveyor.yml`` file.

GNU ``make`` needs to be run from within a ``bash`` shell environment. However,

it is **very** important to use the ``gcc`` and ``gfortran`` that are shipped

with Rtools. Typically in Windows GUI this could be accomplished by opening an

``MSYS2`` shell, running

``export PATH=\C\Rtools\bin:\C\Rtools\mingw_64\bin:$PATH``, then ``make all``.

This project accomplish the same thing using a ``bash`` script.

Once this is completed, the compiled library is avaiable as a downloadable

Artifact from Appveyor. Go to the project's Appveyor site:

https://ci.appveyor.com/project/thequackdaddy/openblasr

Currently, I we are only building drivers for Intel's Haswell, Sandybridge, and

Nehalem architectures. Others could be easily added by just adding to the

appveyor.yml file.

Pick the most recent build that succeeded and select the Artifact tab to

download it.

.. _openblas: http://www.openblas.net/

.. _r-openblas: https://www.github.com/thequackdaddy/r-openblas

.. _R: https://www.r-project.org/

.. _Rtools: https://cran.r-project.org/bin/windows/Rtools/

.. _MSYS2: http://www.msys2.org/

.. _Appveyor: http://appveyor.com/

.. _`blog post`: https://www.avrahamadler.com/r-tips/build-openblas-for-windows-r64/

.. |Appveyor Build Status| image:: https://ci.appveyor.com/api/projects/status/9xm30c47u8hp894h/branch/master?svg=true

   :target: https://ci.appveyor.com/project/thequackdaddy/openblasr/branch/master