gputools
=========

contents
=========
+ notes
+ machines w/ older gpus
+ machines w/o a gpu at all
+ your cuda install is not in /usr/local/cuda
+ mac os x systems
+ windows systems

========
+ notes
========

The usual installation procedure should go something like this:
1. procure a machine with a cuda capable nvidia gpu
2. install nvidia's cuda capable video driver
3. install nvidia's cuda toolkit
4. install this package using R

The package gputools attempts to follow the standard R package installation
procedure. To install the package, you must first install
nvidia's cuda toolkit (preferably version 2.3) available from

http://www.nvidia.com/object/cuda_get.html

You need not use a gpu, to use the package without a cuda capable gpu and
video driver, please skip to the third section 'machines w/o a gpu at all'.

To use a gpu, you will need a cuda capable gpu and a cuda capable video driver.
You can get the video driver from

http://www.nvidia.com/object/cuda_get.html

also.

=========================
+ machines w/ older gpus
=========================

See

http://brainarray.mbni.med.umich.edu/Brainarray/Rgpgpu/#issues

for a table listing compute capabilities of various nvidia cards.
If you have a gpu that has compute capability < 1.3 then
the 'kendall' method of 'gpuCor' and
the functions 'gpuSvmTrain' and 'gpuSvmPredict'
will not function correctly! All other methods should be fine with such a card.

============================
+ machines w/o a gpu at all
============================

If you do not have a cuda capable gpu or just don't wish to use one read
this section. You need not install a cuda capable video driver. You can
still use the package using one of the following commands.

From a terminal:

R CMD INSTALL --configure-args='--enable-emulation' <path-to-package>gputools_<version-of-package>.tar.gz

From inside an R environment:

install.packages(gputools, configure.args = '--enable-emulation')

Please note that you must still install the cuda toolkit.
Device emulation may not work with cuda toolkit version 2.2.
Version 2.3 is strongly recommended for gpu emulation.

The support vector machine functions are disabled in device emulation mode.
They make use of device driver api calls which are not available unless you
are able to install a cuda capable driver. To install a cuda capable drive,
you usually need a cuda capable device.

==============================================
+ your cuda install is not in /usr/local/cuda
==============================================

If the cuda toolkit is not in '/user/local/cuda' then you can specify the
location of the cuda toolkit by either setting the environment variable
'CUDA_HOME' or using one of the following commands.

From a terminal:

R CMD INSTALL --configure-args='--with-cuda-home=<path-to-cuda>' <path-to-package>gputools_<version-of-package>.tar.gz

From inside an R environment:

install.packages(gputools, configure.args = '--with-cuda-home=/usr/local/cuda')

====================
+ mac os x systems
====================

All of the above should work on mac os x systems and even in the R gui without
modification. If something doesn't work as expected on the mac please email
the maintainer.

==================
+ windows systems
==================

There is no support for Windows yet. If you would like to write a windows port,
please feel free. You are welcome to email it to the maintainer for merging
with the main package, but only if you want to.