GPGPU Programming for Games and Science

20 GPGPU Programming for Games and Science

Naturally, it is desirable to design algorithms and implement them to ob-

tain accurate results in a robust manner. Doing so in a timely and efficient

manner may not be easy or even possible.

2.2.3 Speed

Robustness and accuracy do not always come for free. In my computa-

tional geometry code, I provide the ability to compute using exact rational

arithmetic, treating the input floating-point numbers as rational numbers. The

rational arithmetic is based on exact integer arithmetic for integers that have

more bits than supported by standard CPUs. This means that the numerical

computations are performed in software and are quite slow. This is a prob-

lem for a real-time application that must compute convex hulls