Now the CUDA compute capability is not hard-coded anymore.
Setting

--with-gpu-compute-capability=value

set the nvcc compiler option "-arch value".

A simple sanity checking is done to ensure that "value" is not
completely unreasonable

Due to a bug in the Intel compiler some lines could not be
generalised.

Due to bug in Intel compiler some ifdefs are still necessary

Due to a bug in the Intel compiler two stages could not be unified

Now it possible for both the real and complex case to switch on
the blocked code path in the cpu version of band_to_full.
This closes issue #42.

Still to do:
- make blocking a run-time option
- allow to tune the blocking parameters at run-time