CUDA 9.1

[bnemetchek]

Firstly, thanks for this very useful tool!

Your source seems to run well against the 9.1 nvcc toolkit. I see that you had mentioned in your readme that you'd like, one day, to be able run the code against a card to get timing results. That would be helpful but I can imagine would take a bit of work.

I was thinking that it also might be useful to display the clock cycles required for each operation in the PTX or SASS. I was wondering if you are aware of any reference or source for this type of information?