Ray derivatives

[DrBalthar]

Hello

I think I've got a very tricky question here but maybe someone can crunch it since I so far was unable to

I wonder is there any way to access the ray derivatives in a shader? I explicit mean the RAY not surface derivatives. If you wonder why you basically need the ray derivatives for example for super sampling or accurate texture sampling (to estimate the anisotropic sampling direction) and they normally should change under reflection and refraction.

The state object itself doesn't hold that information and I couldn't really find any method to get them either. Does mental ray not make this information public accessible? If yes that would be a real shame.

[pazoo]

Hi,

unfortunately, it seems that mr does not calculate ray derivatives at all (which is a shame...).
What you can do is to overload all the mi_trace_... and store your custom ray derivatives with mi_shaderstate_set
It will only work with your custom shaders though...

[DrBalthar]

Originally Posted by pazoo

Hi,

unfortunately, it seems that mr does not calculate ray derivatives at all (which is a shame...).

Hello pazoo thanks for your answer. Hmm are you sure about this? That would be really strange since mental ray definitely would suffer from texture sampling issues if it doesn't and would lead to very inefficient anisotropic texture sampling (over / undersampling)

What you can do is to overload all the mi_trace_... and store your custom ray derivatives with mi_shaderstate_set
It will only work with your custom shaders though...

Well I am afraid that does not really help much, since I don't really know the initial incoming ones either fom the lens shader or if it is a reflection / refraction ray from a different material.

Oh well guess can't do anything about it.

I wish mental images will add that in one of the upcoming mental ray versions (altough very unlikely).

Thanks again for your answer and confirming my guess.

[pazoo]

You're right, actually it is a real problem for filtering. If you look at the anisotropic filtering example, you will see that they only compute the filtering size using the camera projection matrix (this only works for primary rays, for secondary rays you have to think of something yourself).

And yes, if you compute the ray derivatives yourself you have to go only with your shaders (beginning by a lens shader...)

I've been asking for ray derivatives for years now, but no success so far, maybe if we are many to ask for it...

[DrBalthar]

Originally Posted by pazoo

You're right, actually it is a real problem for filtering. If you look at the anisotropic filtering example, you will see that they only compute the filtering size using the camera projection matrix

Ah good point I forgot to look at that one that would have given me some clue.

Originally Posted by pazoo

And yes, if you compute the ray derivatives yourself you have to go only with your shaders (beginning by a lens shader...)

As I said before that is unfortunately not feasible in my use case scenario.

Originally Posted by pazoo

I've been asking for ray derivatives for years now, but no success so far, maybe if we are many to ask for it...

Well let's hope so I am definitely rallying for it as well now and I will tell all my users to nag mental images to include that feature . So we will see.

[juri]

This functionality (including computation of differentials over reflections and refractions) has been added in mental ray 3.8.

Juri

[MrBios]

Hello I've run into a similar issues with one of my shaders
(http://www.antiflash.net/microfacetglass).

So I saw this post. Juri can you tell where you can get / set this derivative information inside your own material shader?

I checked the state object and shader.h (with mr3.8 headers) and I couldn't obviously see any easy way to get this information.

Thank you!