We slightly modify the baseclass header to contain the number of
blocks and threads that will be used by the kernels. This will prevent
us from having to recalculate these values in each of the GPU routines
again and again. Strictly speaking, these values should be stored in
the DefaultGPU-class, as they're not being used by the CPU
implementation. However, the DefaultGPU-header will be
regenerated when we call HyCuda again for some reason, thereby losing
any changes we made. This is not true for the baseclass header.
Joren Heit
2013-12-17