• Advertisement
Sign in to follow this  

Modify Tessellation shader to make equilateral triangles?

This topic is 1862 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

Continued from: [url="http://www.gamedev.net/topic/635357-quad-tree-lod-but-with-triangles/"]http://www.gamedev.n...with-triangles/[/url]

[b]Details:[/b]
I'm in the proccess of procedural planet generation; so far I'm doing the sphere work.
Currently my sphere is a subdivided icosahedron. (20 sides all equilateral triangles)

Before, when I was subdividing using a software algorithm, one triangle would be
split into four children across the midpoints forming something close to a
sierpinski triangle every iteration...like this: [url="http://puu.sh/1xFIx"]http://puu.sh/1xFIx[/url]

As you can see, each triangle subdivided created more and more equilateral triangles.

But now that I am using the GPU to tessellate in HLSL, the result is definately not
what I am looking for: [url="http://puu.sh/1xFx7"]http://puu.sh/1xFx7[/url]

[b]Questions:[/b]
Is there anything I can do in the Hull and Domain shaders to change the tessellation
so that it subdivides into sets of equilateral triangles like the first image?

Should I be using the geometry shader for something like this? If so, would it be
slower then the tessellator?

Thanks for the time Edited by Dr. Krunklehorn

Share this post


Link to post
Share on other sites
Advertisement
I think I speak in the name of some of the other nearly 249 people when I say that I don't have a good answer to your question. :)

I have done dynamic refinement of triangle meshes before, and one has to be careful to make sure that you don't introduce gaps between triangles where you refined one and not its neighbor. The naive method that divides a triangle into four by subdividing the sides in half would suffer from this. The method of dividing the longest side as long as that side seems too long doesn't suffer from this problem, and it could be what's resulting in the pattern you don't like (I think it looks kind of cool, actually).

Share this post


Link to post
Share on other sites
[quote name='Álvaro' timestamp='1355344204' post='5009950']
I think I speak in the name of some of the other nearly 249 people when I say that I don't have a good answer to your question. [img]http://public.gamedev.net//public/style_emoticons/default/smile.png[/img]
[/quote]

Ah okay that's fine. I was hoping someone else had come across a similar problem with the default tessellation pattern in DX11.

I've done a ton of research since I first posted this question. Articles, PDF's, white papers, tech demos etc. have all helped me
get a better understanding of what the domain shader actually see's and what I have to work with in it.

I'm going to see if I can find a way to do this in the domain shader. (cracks are a fix-later problem) If I can't, that's fine too. I
have an excellent plan to use the geometry shader instead.

What I'd like to be sure of is the performance difference though. If I were to mimic the tessellation shader in the geometry
shader, would the geometry shader method operate slow-er enough to call it a big deal?

[quote name='phantom' timestamp='1355345960' post='5009963']
My thought was why do you require equilateral triangles to be generated?
[/quote]

If I was to subdivide a cube into more and more quads then map that to a radius, the result has an ugly distortion around
the corners. Using equilateral triangles from an icosahedron base nearly solves this problem because it is the highest
order platonic solid you can make. I.E: it is the best "base" to start from if you want to avoid that corner distortion side
effect of the cube method.

Share this post


Link to post
Share on other sites
[quote name='Dr. Krunklehorn' timestamp='1355372739' post='5010066']
I've done a ton of research since I first posted this question. Articles, PDF's, white papers, tech demos etc. have all helped me
get a better understanding of what the domain shader actually see's and what I have to work with in it.

I'm going to see if I can find a way to do this in the domain shader. (cracks are a fix-later problem) If I can't, that's fine too. I
have an excellent plan to use the geometry shader instead.

What I'd like to be sure of is the performance difference though. If I were to mimic the tessellation shader in the geometry
shader, would the geometry shader method operate slow-er enough to call it a big deal?

[/quote]

Have you inspected this PDF?

[url="http://developer.download.nvidia.com/presentations/2009/GDC/GDC09_D3D11Tessellation.pdf"]http://developer.dow...essellation.pdf[/url] [url="http://developer.download.nvidia.com/presentations/2009/GDC/GDC09_D3D11Tessellation.pdf"]http://developer.dow...essellation.pdf[/url]

You could get what you want (terrain height) if you use displacement mapping and keep the equilateral tessellation.

EDIT: Also check out this video --- are these the kinds of triangles you want to end up with? [media]http://www.youtube.com/watch?v=m73UVyKkyGQ[/media]

[url="http://recreationstudios.blogspot.com/2010/03/simple-tessellation-example.html"]http://recreationstudios.blogspot.com/2010/03/simple-tessellation-example.html[/url] Edited by Steve_Segreto

Share this post


Link to post
Share on other sites
[quote name='Steve_Segreto' timestamp='1355379109' post='5010086']are these the kinds of triangles you want to end up with?
[/quote]

That's actually a perfect demonstration of what I am trying to avoid. The method shown in that video is exactly
why cube based spheres get distorted.

Notice the vertex density on the top of the sphere, compared to the vertex density around what would be
the corner of the original cube.

Already there is a big difference: the triangles on top of the sphere are stretched out and the ones near the
original corner point are very small and clustered.

Now if the camera in my planet application was to fly around on ground level, the effect of this distortion
would be easily noticeable. A mountain range around a clustered corner is going to look much better
then a mountain range around the stretched areas on top.

Equilateral triangles is the solution here. They stay the same size all the way around the sphere.

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement