Hi guys,

I have some problems understanding the z-Buffer and how to interpolate the values between vertices correctly in respect to perspective.

What I have done so far is take a vertex in World Space (x, y,z) and transform it using my ViewProjection Matrix to get coordinates in clip-space of the form (x, y, z, w). These coordinates still need to be divided by w.

What do I have to use here z or w? And how do I interpolate it?

What I would do is:

- normalize (x,y,z,w) => (x/w, y/w, z/w, 1)

- linearly interpolate z/w

Would that be correct?

Do I have to divide by 1/w at the end? IF no, why not?

OpenGL is interpolating linearly aswell once you are in clip space and normalized your coordinates afaik. Your coordinates look like this:

(x,y,z,1) where x,y,z are in the range [-1, 1].

Any more opinions on this? My impresison is that I dont have to do perspective correct interpolation on the z here but a linear interpolation is correct since z is already divided by w.

Yep thanks, i am right. You were wrong.

After transforming to the NDC,

So the depth value can be directly interpolated using z-NDC for depth test.

http://www.altdevblogaday.com/2012/04/29/software-rasterizer-part-2/

Conclusion
In this post, the steps to linear interpolate the vertex in screen space is described. And for rasterizing the depth buffer only (e.g. for occlusion), the depth value can be linearly interpolated directly with the z coordinate in NDC space which is even simpler.

However we cannot directly interpolate those attributes in screen space because projection transform after perspective division is not an affine transformation (i.e. after transformation, the mid-point of the line segment is no longer the mid-point), this will result in some distortion and this artifact is even more noticeable when the triangle is large[.]