When would it be more suitable to do Parallax Occlusion Mapping(POM) in a deferred renderer? It seems I can either: During the creation of the G-Buffer phase, I calculate the POM data and store the final result in the G-Buffer. But it seems a waste if they get mask/covered by other geometry in the subsequent draws. Or, during the g-buffer creation I can store the height information in one of the g-buffers channel. If no height data is present, we store a zero value. Then in the first ambient pass, the POM data is computed if there is a height data present, and the result store in the output buffer. Subsequent light-pass do not need to redo the calculations as their result are additively blended to the output buffer. The 2nd method seems to prevent calculation of unnecessary pixels work. Which method is more viable to pursue? Or is there a other solutions? Thanks

I'm using the first technique (not with POM, but CMS, but anyway).

I am not sure the second technique actually works, but maybe I don't understand it correctly. How do you do (any kind of) ray marching if you only have one sample? All other samples in the G-Buffer don't necessarily belong to the same surface or are in the same space. I don't see how you could use them in a valid way.

ahh, you are right, I wasn't thinking hard enough, there will be a lack of heoght samples. Well I just badly wanted to avoid the extra pixel calculations if it is not going to be drawn. Looks like it must have gotten over my head. :D

What exactly is CMS? I don't think you meant 'Content Management System'?

Bad spelling, sorry: CSM, cone step mapping.