I implemented deferred rendering in my little engine using framebuffers but there is no dynamic lighting for now. The g-buffer contains positions and normals in view space and albedo.
To implement real time lighting, I guess that I need to perform one pass for each light. Given the position and shape of a light, how can find all pixels affected by the light? I want to support point lights, spot lights and directional lights.
In deferred shading, one usually just draws some geometry that bounds the light's area of effect in world space, such as a polygonal sphere or cone that encloses the point or spot light's mathematical sphere or cone.
The simplest thing to do is draw the back faces of this geometry (front face culling) with the depth test reversed (greater-than instead of less-equal). Then any pixels that are between the camera and the back faces of the light geometry will get shaded.
This also shades pixels between the camera and the front faces of the light geometry, which are outside the light and therefore don't need to be shaded. There are various optimizations that can be done to remove these pixels, such as a depth bounds test or stencil masking.
For a directional light, you do a full-screen pass.
