View Full Version : Tesselator and geometry shader talk

08-23-2005, 06:07 PM
There is a document released from ATI titled "Shading Compilers" from Avi Bleiweiss.
Anyone read it?
It's not clear to me.

Geometry-pixel shader stages all share the same processing engines and resources. As a result connectivity is much more transparent amongst the shader units and outputs from one can be redirected as input to another almost seamlessly. The introduction of both the tessellation and the geometry shaders with intrinsic expansion properties clearly changes pipe dynamics. This potentially opens up alternative schemes for partitioning load across GPUs. Next, a displaced and motion blurred geometry will be used as a walkthrough example to further illustrate internal pipe interaction.

08-23-2005, 11:51 PM
I hope I understood it.
Anyway, I've thought of it as I'm just discovering shaders. For me, processing a shader can pass threw different units in order to be achieved. I don't know if it will really help things to process faster, but that might help.
I'm sure this will help almost for 'cutting' a process into several more little others that could each be achieved in an efficent manner.
After all, each current card has between 8 and 16 pipes.

Hope that's not too bad...

08-24-2005, 06:01 PM
I think you haven't understood it.
This is about new shaders (tess & geom) and how the graphics pipe will work, scheme-wise. It's not for solving performance related problems within the graphics pipeline itself.
It will be a new feature and it will come with it's own problems, I'm sure.

08-24-2005, 06:21 PM
I think basically he's talking about how there is no more a separate pipeline for vertex shaders and pixel shaders. Every shader, both vertex and pixel goes through the same pipeline hardware. A unified shader pipeline if you will. It also seems that he's talking about new shader types that will allow us to do more things, like our own tesselation techniques, that will run just fine in the unified shader pipeline. So basically you have all of these shaders that run in the same pipeline which allows the outputs of some to be inputs of others.

I may be way off but that's what I thought of when reading that quote.


08-27-2005, 05:00 PM
Something like that. The comment is rather abstract. It would be nice have a few pages on the matter.

08-27-2005, 05:59 PM
Yes, that and maybe someone else explaining it in a different way. :) Honestly I was kind of struggling with that quote myself. The only thing that made any sense to me was the reference to a unified shader pipeline.