
Optimizations
-------------

The ICP code is reasonably optimized (the correspondence search in particular).
But only algorithmic optimization were used. In particular, the following were *not* used:

- look-up tables
- fixed-point math
- chipset-specific extensions for vectorizations (altivec, etc)
- optimizations that would give an approximation to the answer
- assembler in general

All computations done in *double* precision (*float* would be more than enough).
