shader feedback for collision and other non-graphical uses

[valoh]

How do you handle collision/physics with a shader driven architecture?

Wouldn't be some kind of software processing extension be usefull for non-trivial shader vertex transformations?
So that you can use your vertex shader (with some preprocessor directives to leave out non-geometricly calculations) for rendering with CPU to main memory.

Is something like this possible or planed as ARB extension? Or how I'm supposed to handle the feedback of shader transformations? I think for low poly collision geometry and bounding volumes it doesn't make sense to use the GPU and read the result back from graphic card.

[valoh]

no one?

I mean it can't be reasonable to hardcode the transformations or preprocess shaders to get the transformation on you own.

How is this problem supposed to be solved with glslang?

How do other platforms (HLSL, Cg) handle this?

[mcvillar]

I found some work about in Vassilev paper "Fast Cloth Animation on Walking Avatars", he used the z-buffer.

And most recently and related whit your question, Knot paper "Particle System Collision Detection using Graphics Hardware".

The general idea is ... use maps ... depth maps and impact maps, calculated in the GPU.

I'm progamming with Cg some physics things for my Tesis.

Regard,



Originally posted by valoh:
no one?

I mean it can't be reasonable to hardcode the transformations or preprocess shaders to get the transformation on you own.

How is this problem supposed to be solved with glslang?

How do other platforms (HLSL, Cg) handle this?