What can't you do in CUDA that you'd like? Requests for the future

[Cygnus_X1]

I think every compliler parameter could be able to disable/enable/set through #pragma per each kernel. My maxregcount, your MAD, but in future some other parameter may go into someone’s way…