A limiting factor in the performance of a rendering system is the number of state changes, i.e., changes of the attributes material, texture, shader program, etc., in the stream of rendered primitives. We propose to include a small buffer between application and graphics hardware in the rendering system. This pipeline buffer is used to rearrange the incoming sequence of primitives on-line and locally in such a way that the number of state changes is minimized. This method is generic; it can be easily integrated into existing rendering systems.
In our experiments a pipeline buffer reduces the number of state changes by an order of magnitude and achieves almost the same rendering time as an optimal, i.e., presorted, sequence without pipeline buffer. Due to its simple structure and its low memory requirements this method can easily be implemented in software or even hardware.