GLSlang Overview

[Dave Baumann]

If anyone's interested:

http://www.3dlabs.com/support/developer/ogl2/presentations/GLslang_GDC2003.pdf

 

[demalion]

Looks good, but some of the specifications seem to preclude (to my understanding) some of the functionality that would be allowed by DX PS/VS 3.0....namely, the vertex can take texture memory as an input, but all the data flow connections I noticed seem to preclude the direct fragment shader output to vertex shader input I thought was allowed by some DX VS/PS 3.0 registers.

Perhaps the compilation could take advantage of the functionality in the hardware...or those enumerated "user-defined attributes" might allow such a concept to be expressed?

 

[Simon F]

Dave, thanks for the link. I've obly had a brief glance but it looks quite interesting.

...but all the data flow connections I noticed seem to preclude the direct fragment shader output to vertex shader input I thought was allowed by some DX VS/PS 3.0 registers....

I don't think there is any such 'direct' connection in DX either.

 

[ultrafly]

In XBOX,use XGSetVertexBufferHeader to set Texture as Vertex Buffer.
Use this method,you can render to vertex buffer(use the same data with the texture).

 

[Simon F]

In XBOX,use XGSetVertexBufferHeader to set Texture as Vertex Buffer.
Use this method,you can render to vertex buffer(use the same data with the texture).

I don't think you could call that a direct connection - you'd surely have to do some sort of gross, texture/rendering-level synchronisation.

 

[demalion]

I had thought I had seen something indicating that the output of a pixel shader op could be used as an input for vertex shader under PS/VS 3.0, without an intermediary "-> texture ->", and that's what I meant. The impression I have is of something stating this clearly, but I can't seem to find it or recognize the statement again.

I guess I was imagining it?

 

[Humus]

I can't get direct output from a pixel shader to a vertex shader to make much sense.

 

[demalion]

I can't get direct output from a pixel shader to a vertex shader to make much sense.

Hmm....I guess if the vertex shader has all the pixel shading instructions that might conceivably be useful for vertex manipulation, it doesn't, unless the pixel shader pipelines are laying idle at the time. Which has to do with the assumption I tend to make about where nvidia is heading with the nv3x (and the questions about pixel/vertex shader performance interactions for the cards I still have) and my interpretations of the R400 rumors.


Anyways, which pixel shader ops might be useful for that, but aren't in the VS 3.0 specification? If there aren't any, then the issue is simply one of computational efficiency, rather than preventing bandwidth bottlenecking (which is what pixel shader -> texture -> vertex shader seems like it will be doing).