Patryk Kizny Raytracer Updates

[lycium]

Friends don't let friends use that sin-thing It is 100% certainly causing tons of errors everywhere it's used.

The Wang hash, while infinitely better than the sin-thing in all ways, still is not a good uniform random number generator (it's not even a random number generator, just a uniform-ish hash function)... if you're doing Monte Carlo, you want something accurate and high quality, else additional samples are more damaging the image than improving it (by injecting more and more influence of the b0rked function).

[Patryk Kizny]

What can you suggest then?
I posted a bunch of noise functions here: http://www.fractalforums.com/index.php?topic=22522.0
But these are usually pretty expensive.

[lycium]

Erm, those are something totally different though: those are noise functions, continuous functions like Perlin noise.

What's needed here is a (uniform) random number generator, or RNG. RNG (or tables) is needed for things like Perlin noise, not vice versa.

[Patryk Kizny]

Oh, right. Forgot about that.

[lycium]

I've probably said it too often by now, but i still think it's somewhat a shame that we have kinda shared goals but work separately.

I'm in Wroclaw from friday to tuesday, in case you want to maybe meet up; might end up living there a few months even.

[Patryk Kizny]

I've probably said it too often by now, but i still think it's somewhat a shame that we have kinda shared goals but work separately.

I'm in Wroclaw from friday to tuesday, in case you want to maybe meet up; might end up living there a few months even.

We have a great party here on Friday evening - come over.
https://www.facebook.com/events/555370944644986/

[phtolo]

Here's an example rng I have used in Framentarium.

Code:

int rand(in out ivec2 prnp)
{
	int	r0 = prnp.y >> 15;
	int r1 = prnp.x ^ r0;
	r0 = r1 << 17;
	prnp.y = r0 ^ r1;
	prnp.x = 69069 * prnp.x;
	r1 = prnp.x ^ prnp.y;
	return r1;
}

Inside your main function, create a seeded variable to keep the current state:

Code:

ivec2 prnp = ivec2(1257353,135);

Then call rand with the current state-variable to get a random value:

Code:

int irnd = rand(prnp);
float frnd = float(rand(prnp) & 0xfffffff)/268435455.0;

[Crist-JRoger]

Here's an example rng I have used in Framentarium.

  I copied your code into calculating disk for Dof. And nothing ) image is clear, Dof don't works absolutely. Anyway wanghash works!

The Wang hash, while infinitely better than the sin-thing in all ways, still is not a good uniform random number generator (it's not even a random number generator, just a uniform-ish hash function)... if you're doing Monte Carlo, you want something accurate and high quality, else additional samples are more damaging the image than improving it (by injecting more and more influence of the b0rked function).

What is better? 
Found this post on 1 google page )
With wang hash "holey" bokeh works well. Thank you!

[Patryk Kizny]

Need help. I guess it's a basic thing, but can't get my head around that.

- I have implemented a multi material thing in the raytracer.
- say therea are fractal objects with different materials, and two DEs;

Code:

m1, m2, de1, de2

- there's a function to blend between these materials:

Code:

blendMat(m1, m2, ratio)

working more less like mix(f1, fr, r) but on structs.

final DE is computed using IQ's sminpol function:

Code:

// polynomial smooth_ min (k = 0.1); --> fastest
float sminpol(float a, float b, float k) {
    float h = clamp(0.5 + 0.5 * (b - a) / k, 0.0, 1.0);
    return mix(b, a, h) - k * h * (1.0 - h);
}

as follows:

Code:

DE = sminpol(de1, de2, k);

Now as far as I understand based on documentation: http://www.iquilezles.org/www/articles/smin/smin.htm
sminpol returns values that are equal to min(de1, de2) or smaller.

Now what I need is to find a ratio that would blend materials. I don't need anything fancy, linear interpolation is fine.
It should give me:
0 if distance from DE to de1 == 0
1 if distance from DE to de2 == 0
[0 ... 1] for the smoothed corners.

How?

[eiffie]

Have you tried just using the value "h" from IQ's formula to mix the materials?

[Patryk Kizny]

Here are two options:

[01] Blending with k as you suggested:
[02] Blend with the following formula:

Code:

float br = (min(de1, de2) / max(de1, de2)); // this gives 1 at the joints and 0 at non-smoothed areas
float m = de1<de2 ? 1 : 0; // comparing DE to check which one was closer if it was not blended
matBlendRatio = mix(m, 0.5, br); // blend between original object color and perfectly blended color (1/2 ratio) on joints

This looks more less good but I believe it is not an elegant solution. Someone with a better math skills could probably come with a simpler and way more elegant mapping function.

// edit
Actually, when I used separate K for color blending and a separate one for geometry this gives me good control. Solved, thanks!

[Crist-JRoger]

Stupid quesion, but as user I dare to ask..
Is there other way to render 3D objects, faster and easy than DE(I am confused in terms of, in this case I am referring to the use of a raysteps) and the same time having similar characteristics? Just spend some time at shadertoy site, some shaders works well even on slow Intel GPU. Than with Eiffie's help i moved one shader to DE-render and it has become slow...

[eiffie]

If you use 3D.frag to set up the camera it does multi-sampling. The shaders on ShaderToy typically do not use multi-sampling so they are faster.
To answer your question (is there a faster method than DE) yes just turn the DE into a mesh and draw it as millions of polygons. Look at other threads for information on how to do that.