Re: [x3d-public] Shaders and Triangles: was dedicated working-group focus on X3D interoperability

["Holger Grahn" <hg@snafu.de>]

 

Hi John

 

Regarding the Triangle nodes I agree that X3D has sometimes a bit of a node bloat philosophy.

 

The IndexedFaceSet could represent handle it all.

 

On the other hand IFS in its generic form with per face attributes,  creaseAngle and seperate index list into attributes array

doesn't  fit the Open GL or Directx layout.

To render an IFS, one need to reformat/copy the data to flattened array structures.

 

The Triangle Set nodes are more constrained, they are fitting better direct the OpenGL geometry mapping.

But  for DirectX9 requireing the interleaving of all attributes of a vertex into one structure, they don't help much.

 

Hardware Shaders so far operate indepentently on vertices and pixel / fragments, they can work on all primitives,

I don't see Justins point here, beside there might be differences how implementations compute normals etc.

 

This changes with DirectX 10 adding geometry shader capability where the actual input triangle structure is important.

Unfortunately in DirectX10 you always need to implement all appearance using a shader,

so the classical OpenGL lighting/material/texture/multitexture/fog  model has been completly removed.

 

Greetings

 

Holger

 

 

 

 

 

----- Original Message -----

From: John A. Stewart

To: Justin Couch

Cc: X3D Graphics public mailing list ; Web3D Consortium Members

Sent: Wednesday, December 20, 2006 9:10 PM

Subject: [x3d-public] Shaders and Triangles: was dedicated working-group focus on X3D interoperability

Justin;

Throwing in some more canadian pennies in the bowl;

I'll go one step further. It's impossible to use Shaders without the Triangle* nodes. Shaders require direct knowledge of the exact geometry structure coming at them as well as precise control over every aspect of the incoming geometry in order to work. Remove the Triangle nodes and you must also remove Shaders too as they're pointless without them.

We still don't need the *Triangle nodes for shaders. What the browser would

need to recognize in your example, is that it should *not* optimize, and feed

the triangles as encountered.  *that* mapping can be specified, and the browser

certainly would know if a Shader was present, or not. 

While shaders are certainly cool, and my GPUGems books reside within an

elbows reach of me, I doubt whether many on this list would choose to, um,

teach shader programming in high school, for instance.

Shaders are neat, and have their place, but, at the moment, their application

is rather specialized.