Welcome to the 'WideLM' package.

A few notes which may help with installation:

 - There is currently no version checking in the configuration script,
   so the following should be verified independently:

    - It is assumed that a working CUDA card, with compute capability
      >= 2.0, has been installed.

    - It is also assumed that the CUDA Toolkit has been installed, with
      version >= 4.1.  Runtime problems have been noted with older
      versions.

 - The installer will permit specification of the CUDA home directory
   by means of the '--with-cuda-home=[CUDA home]' option.

 - The CUDA compiler, 'nvcc', uses the resident 'gcc' to generate code
   for the CPU.  Version 4.1 of the Toolkit offers an 'nvcc' which
   will not work with 'gcc' versions more recent than 4.5

   If necessary, alternate versions of 'gcc' can be installed on the
   host system.  In order for the package installer to use an alternate
   version of 'gcc', the '--with-compiler-bindir=[alternate GCC path]'
   should be used.


Apologia:

  This is a first release.  We have done our best to weed out bugs,
  but they may well still be lurking.  Please notify the maintainers
  of any bugs encountered.

  The current version is limited to two predictors, plus interaction.
  It should be possible to compute general-order interactions on the
  GPU, but this is tabled for a future release.

  The code contains architecture-specific constants, such as a maximum
  thread count, which preclude use on compute-capability 1.3.  We
  may be able to clean these up at some point if there is enough
  interest.

Kudos:

  Special thanks to Josh Buckner, lead author of the 'Gputools'
  package, who helped make this code CRAN-friendly.

  Thanks also to Dirk Eddelbuettel, coauthor of the 'Rcpp' package,
  who also helped get this package off the ground.

  
