Lost in the triangles moved! find it here: aras-p.info/blog

I just can't wait to write geometry shaders :).
# posted by  Anonymous : 02:18  

Yeah, GS is a very interesting piece...

Though I can't really imagine how the hardware will be able to pipeline everything efficiently (when each input primitive can end up with variable number of primitives).

The possibility to do this is great, but I think most of the interesting scenarios will be pretty slow as well. Of course, "slow" is a relative term (i.e. it'll probably be faster than the alternative methods of doing the same thing)
# posted by  NeARAZ : 10:56  

Well that works pretty well for PS2 :) ('well' is a relative term of course)

I suppose gpu already has very deep pipelines, plus you will specify your maximum amount of vertices statically - so gpu can schedule its work ahead pretty well.

And probably :) it's not impossible for gpu to have more than one independent vertex pipe running at the same time.

Or VS/GS could be linked into the one uber-shader underneath (much like PS2 vu1 code) and executed in the same block - could be no worries about the VS stalls then.

Kris: no need to wait - go write vu code :)
# posted by  ReJ : 20:55  

ReJ: if VS+GS would end up as a single piece of hardware, then why would it be exposed as two separate stages?

Analogy with PS2 makes sense though :)
# posted by  NeARAZ : 21:18  

NeAraz: why separate VS/GS, let's do some wild guessing here:
a) easier to program I suppose, backward compatible in some sense (easier to reuse legacy vshaders)
b) maybe easier for compiler to schedule
c) more options for underlying hardware - could be one block, or could be separate blocks, but with 'multithreading' to hide stalls
d) ... something really important
# posted by  ReJ : 11:14  

ReJ: I'd rather wait for a first dx10 card. Besides, now I have one game project at work, one at home and university :).
# posted by  Anonymous : 13:31