[cairo] Some performance investigations

[chris nuernberger]

Are you sticking with the 128 fixed point math?

I looked through the 128 bit division, and it should be relatively
easy to do higher-radix math (4 or eight would probably work.  Any
higher than that and the setup and if statements required would
probably kill you).  Higher radix math would look a lot like just
straight loop unrolling, so perhaps you could just do that instead.

I don't use straight C very much, so I was wondering if perhaps part
of the problem was function call overhead?  Or is the compiler smart
enough that since it can see the entire function definition it will
inline stuff where possible?

Chris