use isl for CUDA code generation
Since CLooG and the isl code generator are quite different,
especially in the way they handle nested code generations,
the switch to the isl code generator results in quite
a few changes throughout the code.
In particular, where the original code would have to pass quite
a bit of information through parameters (hoping that they wouldn't
get reordered), this information can now be passed through the schedule,
resulting in significant simplifications throughout.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>