INSTALL

The following are required to use and install this package:

 - An NVIDIA GPU (!)
 - The CUDA SDK (see http://www.nvidia.com/object/cuda_home_new.html)
   which includes the NVCC compiler
 - SLang (www.jedsoft.org/slang/)

With these components installation should be simple. The default
makefile assumes the SDK components are in the directory

/usr/local/cuda/*

Just edit as necessary, and build and install with:

make
make install

GPUs come in different compute capabilities, which can only be
detected once code is compiled. This is specified with the -arch flag
in NVFLAGS. Default is 1.0 which is the lowest capability and does not
have double support. If you do not know the capability of your GPU,
simply compile once, then in a slsh shell do

slsh> require("cuda");
slsh> i=cuda_info(); % the default device
slsh> ()=printf("%d%d\n",i.major,i.minor);
30

That number prefixed with "sm_" can be used for the arch to get the
best performance out of your card.