Fix disk-blending of per-view meshes on OpenGL

[florian-schweiger]

Fixes problems that prevent correct disk-based blending on OpenGL platforms.

Direct3D and OpenGL use different conventions for the range of the z-coordinate in normalised device coordinates as well as for the encoding of depth buffer values (see Platform-specific rendering differences).

In particular, the NDC z-range is [0,1] and [-1,1] in D3D and OpenGL, respectively. Also, raw depth buffer values in D3D are reversed, i.e. values range from 1(=near) to 0(=far), whereas OpenGL uses 0(=near) to 1(=far).

While many Unity macros and shader helper functions seem to abstract these differences away, this does cause problems in certain places. Notably on OpenGL platforms, UnityObjectToClipPos yields negative z-values close to the near clipping plane, and sampling high-precision depth textures produces raw values that increase with depth.

This commit adds checks to make NDC z-scaling and depth testing work equally on OpenGL and Direct3D.

[DinechinGreg]

Very well noticed! I had tried to adapt the shader code for these platform-specific differences, but had not seen that it became an issue for disk-based blending, so this is definitely an important fix. Thank you very much for the commit!

[florian-schweiger]

It is quite confusing, and Unity's docs are not very clear on this either, unfortunately.

I understand that tex2D look-ups inevitably return platform specific values, but I'm somewhat disappointed that Unity's functions/macros are not consistent in their behaviour (as far as I can tell).

There are probably other parts in the code that are affected by this. Disk blending is the only one I had a closer look at.