I am looking the cuda programming guide.
In the examples of it, they use float for datas, not int.
I wonder that there is some advantage when I use float type instead of int type :)
You can store fractional values in them and they are less likely to overflow.
In terms of performance, floats used to have an advantage in G80. But in Fermi there is not much difference any more, apart from the fact that some integer bit-wise operations still do not have full throughput.