Early Z Rejection

http://software.intel.com/en-us/vcsource/samples/early-z-rejection

This sample demonstrates two ways to take advantage of early Z rejection. When rendering, if the hardware detects that after performing vertex shading a fragment will fail the depth test, it can avoid the cost of executing the pixel shader and reject the fragment. To best take advantage of this feature, it is necessary to maximize the number of fragments that can be rejected because of depth testing. This sample demonstrates two ways of doing this: front to back rendering and z pre-pass.

Front to back rendering requires sorting opaque geometry based on distance from the camera and rendering the geometry closest to the camera first. This rendering order will maximize depth test failure without additional D3D API calls.

For Z pre-pass, all opaque geometry is rendered in two passes. The first pass populates the Z buffer with depth values from all opaque geometry. A null pixel shader is used and the color buffer is not updated. For this first pass, only simple vertex shading is performed so unnecessary constant buffer updates and vertex-layout data should be avoided. For the second pass, the geometry is resubmitted with Z writes disabled but Z testing on and full vertex and pixel shading is performed. The graphics hardware takes advantage of Early-Z to avoid performing pixel shading on geometry that is not visible.

Performing a Z pre-pass can be a significant gain when per pixel costs are the dominant bottleneck and overdraw is significant. The increased draw call, vertex and Z related per pixel costs of the pre-pass may be such that this is not a suitable performance optimization in some scenarios.

PS：如果在Pixel Shader中使用了clip()函数，则Early Z会失效。

参考：http://xboxforums.create.msdn.com/forums/p/99613/654207.aspx