Computation Culling with Explicit Early-Z and Dynamic Flow Control

Prior to execution of the pixel shader, the GPU performs a check of the interpolated z value against the z value in the z buffer. This occurs for any pixels which passed the hierarchical z test and which are actually going to use the primitive’s interpolated z (rather than compute z in the pixel shader itself). This additional check provides not only an added efficiency win when using long, costly pixel shaders, but also provides a form of pixel-level control flow in specific situations. In a number of applications such as volume rendering, skin shading and fluid simulation, the z buffer can be thought of as containing condition codes governing the execution of expensive pixel shaders. Inserting inexpensive rendering passes whose only job is to appropriately set the “condition codes” for subsequent expensive rendering passes can increase the performance of several applications.