Tessellator / PPP Still "Optional" In WGF2.0

[KimB]

Errr, that may or may not be plausible. We don't yet know the instruction set of the geometry shader. But at the very least it'd be incredibly inconvenient, and would prevent any sort of dynamic LOD adjustment.

 

[Ailuros]

Couldn't an IHV use a well designed PPP to cater for both topology and tesselation at the same time?

Granted the risk would be somewhat high, if the tesselation unit is optional; if hypothetically one major IHV has a tesselation unit (in whatever form) and the other one hasn't, then I figure developers will most likely again opt for the lowest common denominator.

Reading though carefully through the Peddie presentation and NVIDIA's plans for the professional/render farm markets, I wouldn't be surprised if NVIDIA would take above risk. Ok there's too much speculation involved, but I really want to see "real" advanced HOS support in the foreseeable future *sigh*

 

[Xmas]

Anyone care to speculate, how one could place a GS in a pipeline with unified PS/VS units?

Simply:

GS--->Shader Units?

That's implying the GS has to be separate

 

[Cyberdigitus]

IMO this is one of the coolest and most important things to come, definitely if you can pass those gpu-created vertices and face indices trough the pipeline in various ways.

shadow volume extrusion, fur/fin, more control over edges and point sprites, maybe even polygonalisation (think marching cubes on gpu), etc

 

[Ailuros]

Anyone care to speculate, how one could place a GS in a pipeline with unified PS/VS units?

Simply:

GS--->Shader Units?

That's implying the GS has to be separate

Ok now I'm really confused:

AFAIK the Geometry Shader doesn't necessarily have the same functonality / instructions as the VS/PS; I believe the only elements that are confirmed as unified (in their instruction set / capability requirements) are the VS and PS.

http://www.beyond3d.com/forum/viewtopic.php?t=19470&postdays=0&postorder=asc&start=0

No wonder I'm asking for clarifications....

 

[Xmas]

I'm not referring to actual hardware, I'm referring to the theoretical possibility of a unified PS/VS/GS. What does a GS need? Fast vector math. And special instructions to output primitives to the triangle setup.

But if it is a separate unit, you need a unified shader -> GS -> triangle setup -> unified shader loop.

 

[Dave Baumann]

Or just a memory bus that can shove this data around fairly efficiently...

 

[MuFu]

Wow, not enough wink smileys in the world for that one.

 

[Mate Kovacs]

What does a GS need? Fast vector math.

IMO heavily SIMD-like integer operations are also crucial (for topology processing).

 

[Xmas]

What does a GS need? Fast vector math.

IMO heavily SIMD-like integer operations are also crucial (for topology processing).

WGF 2.0 requires integer math in PS/VS, too.

 

[Mate Kovacs]

With bitwise ops included? I've always wondered, when would we see the first "good old xor-plasm" shader.

 

[Xmas]

With bitwise ops included?

Yes.

 

[db]

No wonder I'm asking for clarifications....

does looking at the winhec slides help?
http://download.microsoft.com/downl...41f2-893d-a6f2295b40c8/TW04079_WINHEC2004.ppt

I think terms like "unified shaders" and "geometry shader" (as opposed to "unified ps & vs") strongly suggest that all shaders have unified programming interfaces and ihvs are on there own to decide how to best implement those unified interfaces.

of course, those slides are now > 8 months old, so some things may have changed.

 

[Ailuros]

But if it is a separate unit, you need a unified shader -> GS -> triangle setup -> unified shader loop.

Clear enough. Danke

 

[Dave B(TotalVR)]

When is all this gonna be in my PC?

 

[Ailuros]

When is all this gonna be in my PC?

=/>2006.

 

[Randell]

When is all this gonna be in my PC?

=/>2006.

good coz I can't afford an upgrade this year

 

[DemoCoder]

A geometry shader/tesselator would also need random access to data structures too, else it would be severely limited in the kinds of algorithms you could program on it.