Author:  Matthew Curry
Email:   [email protected]

This is a library for Reed-Solomon coding that is designed to be used
with an NVIDIA GPU, preferably with compute capability 1.3 or later.
It includes two sample programs found in the examples directory.

To build the GPU version, simply type "make cuda=1".  To build a CPU
version, type "make cpu=1".  To use Jerasure for coding, which is
maintained by James Plank of UTK, extract it into the "lib"
subdirectory, build it there, and run "make jerasure=1".  See
http://www.cs.utk.edu/~plank/plank/papers/CS-08-627.html to obtain a
copy of Jerasure.

To build the GPU version, two environment variables need to be passed:
- CUDA_INC_DIR: This contains the CUDA include files.
- CUDA_LIB_DIR: This contains the CUDA library files.

To run the GPU version, two environment variables have to be defined:

- GIB_CACHE_DIR: This is where compiled GPU code goes.  This directory
  should not be publicly accessible and should exist.  A typical value
  is the "ptx" directory within Gibraltar's directory.

- GIB_SRC_DIR: This is where the source to the kernel lives.  The
  typical value of this is the "src" directory within Gibraltar's
  directory.

Ensure that, when running Gibraltar applications, the compiler and
libraries for the CUDA version used are available to the application.

Optionally, if more than one CUDA GPU exists on the system, use
"GIB_GPU_ID" to tell it which to use.  Use a number between 0 and
MAX-1.

By default, a compute context 1.3 or greater GPU is assumed.  If this
is not the case, define GIB_USE_MMAP to be 0.