PDA

View Full Version : Question on GL_DEPTH_CLAMP_NV and ATI support or similar extension



Mars_999
08-18-2006, 04:57 PM
Does ATI have a similar extension or have a way to do the same thing as GL_DEPTH_CLAMP_NV? Thanks

cass
08-19-2006, 06:21 PM
ATI would need to answer that authoritatively, but an architecture needs to be able to disable near and far view frustum clipping and clamp sample depth efficiently to do this.

It's not obvious to me that support for this would fall out of many designs.

Mars_999
08-19-2006, 06:45 PM
I just started using this and like it for not clipping my meshes I am rendering close to my near plane... I thought I seen somewhere someone was doing this another way to allow both NV, ATI to run their code... with the same behavior.

plasmonster
08-19-2006, 06:58 PM
GL_DEPTH_CLAMP_NV is really great for stencil shadows too :)

plasmonster
08-19-2006, 07:07 PM
ATI would need to answer that authoritatively, but an architecture needs to be able to disable near and far view frustum clipping and clamp sample depth efficiently to do this.Thanks, Cass. That's making a lot more sense now, given what you said earlier about the possibility that an implementation might not even use geometric clip planes, and instead use guard band clipping, or even avoid clipping altogether.

Humus
08-20-2006, 03:29 PM
We don't support that extension or anything similar. I'm not sure if the hardware could do it.

Mars_999
08-20-2006, 09:49 PM
Originally posted by Humus:
We don't support that extension or anything similar. I'm not sure if the hardware could do it. :(

Jan
08-21-2006, 01:11 AM
I always thought ATI "couldn't" support it, because nVidia holds the rights for this extension.

Overmind
08-21-2006, 01:20 AM
That wouldn't prevent ATI from making a similar extension.

At least for the far plane, the problem can be solved by using "infinite view frustums":
http://www.gamasutra.com/features/20021011/lengyel_02.htm

cass
08-21-2006, 08:57 PM
Yes, when I was originally researching shadow volumes, I used depth clamp to get the effect of an infinite view frustum.

Shortly thereafter, I realized that you could take the limit as f->infinity for the standard OpenGL projection matrix and still get a quite reasonable result. That's what led to the paper that Mark Kilgard and I wrote on it.

There are other reasons why depth clamp on the near plane is nice. It increases the number of cases when you can use zpass SSV rendering and simplifies the math for determining a near plane intersection.

Rendering zpass is usually significantly better if you can do it.

Anyway, I've run off-topic, I fear. :)

Mars_999
08-22-2006, 01:07 AM
Originally posted by cass:

Anyway, I've run off-topic, I fear. :) Never, its a good thing to see someone from Nvidia or ATI comment on these issues, when one is looking for answers. I plan on using the extension for Nivida in my game engine, but not sure what to do for ATI. Wish they would support this would be nice...

plasmonster
08-22-2006, 01:18 AM
That's what led to the paper that Mark Kilgard and I wrote on it.And a nice paper it was :)

When you guys came out with the 2 sided stencil test and depth clamp, it made a lot of SSV lovers very happy.

Now all we need is a NV_really_fast_and_really_soft_ssv extension (he muses with forlorn resignation).