Parallax Occlusion Mapping

[cannon]

https://www.youtube.com/watch?v=PFHD2le0PoM

[cannon]

It's parallax occlusion mapping, not the same thing as that issue.

[Kefta]

Relevant sources:

• parallaxtest.cpp / parallaxtest_vs30.fxc / parallaxtest_ps30.fxc
• materialsystem_config.h / worldtwotextureblend.cpp (addition to material system)
• common_ps_fxc.h (implementation in the standard pixel shader)
• eye_refract_helper.h / eye_refract_helper.cpp / eye_refract.cpp / eye_refract_ps2x.fxc (implementation in eye refraction)
• lightmappedgeneric_ps2_3_x.h / lightmappedgeneric_dx9.cpp / lightmappedgeneric_vs20.fxc / lightmappedgeneric_dx9_helper.h / lightmappedgeneric_dx9_helper.cpp (implementation in LightMappedGeneric, see the commented out code in lightmappedgeneric_dx9_helper, as well)

[cannon]

That POM code is pretty basic. It should: Do a binary search after the initial stepwise search. Apply a random first step offset to noise out the "slicing" artifact. Optionally output an adjusted fragment depth so POM surfaces can actually mesh with geometry. Stepwise search can be optimized by sampling multiple spots along the ray simultaneously then checking them all, so execution doesn't have to be suspended between each step (the HLSL compiler might do that automatically though)

Also note that POM is pretty expensive in general, and should not be run on "low" "effect detail" or whatever it's called

http://www.diva-portal.org/smash/get/diva2:831762/FULLTEXT01.pdf

Just make it so we can write our own pixel shaders :)

[xoticdev]

@Kefta Thanks for correcting my mistake.

Yeah, I definitely agree we should have parallax mapping! It's not really important, but it would give verity to servers.