PDA

View Full Version : Noisefunctions - Will they ever be supported in current HW?



PanzerSchreck
12-27-2003, 12:55 AM
As you can guess, I'm right now dabbling with glSlang, and I started "converting" some of the shaders from ATIs Ashli into my own app.
But it comes with many Fragment Shaders that use noise-functions and they aren't supported in HW on my Radeon9700 atm. And running a fragment shader in SW isn't that usefull anyway http://www.opengl.org/discussion_boards/ubb/wink.gif

So is there any chance that those noise-functions will ever be supported on current hardware, or that they'll be speed up by a newer driver-version?

V-man
12-27-2003, 03:54 PM
I run a few of them (not all).
If a program exceeds hw limits, it does multipass.
Are you saying you are getting <1 FPS?
If so, which one exactly?

Also, is it taking a long time for the demoes to init on your machine?
Looking at task manager, I see it is using quite a bit of RAM.

PanzerSchreck
12-28-2003, 01:46 AM
Multipass is only done by Ashli itself, as it splits glsl-Shaders wenn converting to ARB_VP/FP. But I'm not using Ashli, I'm taking it's shaders, alter them according to the current languagespecs and use them in my own application. And when I use a noise-function there, after linking I'm told that it's an unsupported languagefeature.

That means that noise-functions are done in software, and that's far too slow (<1 FPS) if it's done per Fragment.

So my question has nothing do to with Ashli itself. I just wanted to know if current HW is capable of doing noisefunctions in hardware.

BTW : I've also noticed that gl_FrontFacing (Fragment Shader) is an "unsupported language feature"...hope that changes in the next Catalyst-Release.

[This message has been edited by PanzerSchreck (edited 12-28-2003).]

[This message has been edited by PanzerSchreck (edited 12-28-2003).]

harsman
12-30-2003, 03:54 AM
Well, one way to suport noise functions is to use a lookup in a 3d noise texture. That's certainly feasible on current hw.

PanzerSchreck
12-30-2003, 04:21 AM
I know that it can be "replaced" by this 3D-texture lookup, but it would be much more flexible to just use glSlang's noisefunctions directly in the shader.
And besides this, doing it via a 3D-texture lookup means one less texture lookup left for the rest of the shader.

But the initial idea behind my question was more to know if the current HW (R300 and alike) is able to do noise in HW or not.