Messing with Z

[emackey]

Proposed fix for #2885! Someone with gl_FragDepthEXT experience should review this, as I actually haven't messed with that extension before.

The basic idea is to push the Z values of the fringes of points, since they have low alpha values, towards the back of the current frustum. That way, nearby objects have a better chance of winning any Z battles with those fringes.

I haven't done extensive testing on this, actually I just thought of it about an hour ago. If it seems to work well, we might look for additional uses, for example the fringes around labels.

[pjcozzi]

Do you have a before/after screenshot? It would be interesting to see large points together with 3D models or primitives where the 3D models end up occluding the point fringes, but not the point itself. This part of the point will look aliased, correct? So we are just trading one problem for another with a fudge factor? Could be OK if the new problem isn't as bad. It might also be instructive to try this with billboards/labels to see what the artifacts look like.

The only other comments I have is if this needs to clamp to the far plane and what the best function is (and why-ish!).

Also, this will disable early-z, which could have a significant performance impact on fragment-bound scenarios, e.g., lots of (big) points in a single collection.

[emackey]

We'll find a different solution.