The vector_math library now has less source code as I heavily use the preprocessor to duplicate many parts witch are very similar to each other. It now also has a multiply_accumulate function which many other functions are based uppon. This allows me to specialize this function for fixed point numbers, where a faster version can be supplied and therefore boosting performance. But note that there are still functions which are not particularly fast (matrix inverse, adjoint, determinant). These may be speed up in the future.

The fixed point library now is wrapped into the namespace fixedpoint and provides a specialization for multiply_accumulate. My GP2X demo runs some 10-20% faster with this library and raw vertex transformation speed is more than 25-30% faster.