We present a new and simple data-driven technique for modeling 3D brushes for use in realistic painting programs. Our technique simplifies and accelerates simulation of the constrained dynamics of brushes by using a small lookup table that efficiently encodes the range of feasible constrained states. The result is a brush model which runs an order of magnitude faster than previous physics-based methods, while at the same time delivering greater deformation fidelity.

Previous attempts at vectorizing programs written in a sequential high level language focused on converting control dependences to data dependences using a mechanism known as IF-conversion. After IF-conversion vector optimizations are performed on a data dependence graph. However, IF-conversion is an irrevocable process which can introduce high run-time overhead if the input program is not amenable to vectorization.