Iterated OpenCL Meshes

toneburst's picture

OpenCL experiment. OpenCL meshes being distorted/displaced inside an Iterator.

a|x http://machinesdontcare.wordpress.com

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.

gtoledo3's picture
Re: Iterated OpenCL Meshes

Are you enabling multisampling in the viewer or doing your classic 2x?

Do you think it's better than the old setup, or just cool because it can be done?

toneburst's picture
Re: Iterated OpenCL Meshes

That's with multisampling turned on in the Viewer. I haven't tried my old method, with this one, but I suspect it's slower. Quality is comparable, I think.

a|x

cwright's picture
Re: Iterated OpenCL Meshes

Performance-wise, multisampling will almost always be faster (it doesn't have to deal with downsampling regions of the output that don't need it, etc). It can also be slightly higher quality (the 2x method renders 4 pixels per output pixel, whereas multisampling can potentially render much more (some cards do 8 or even 16 -- my nvidia only does 4 though). I'm guessing QC only does 4 though, so it'll be pretty similar.

gtoledo3's picture
Re: Iterated OpenCL Meshes

One plus of the manual route is that shadows will get anti-aliasing. It's weird to see a smooth edge, and then the shadow associated with it have an aliased edge :)

In this case, it doesn't quite apply, since the shadow would make everything besides the first iteration disappear :/

I was wondering where the multisampling was "fixed at", since one can still see aliasing. 4x sounds about right, looking at the visual result... interesting that it just bumps up to the max whenever enabled.

cwright's picture
Re: Iterated OpenCL Meshes

yeah, the shadow thing sucks (on several fronts) -- not going to deal with that for now ;)

MS technically never "fixes" aliasing -- since the screen is fundamentally bandwidth limited (composed of discreet, finite elements), you'll always be able to spot artifacts under some situations. It simply makes it less apparent. 4x is the basic level of AA, 16x and higher are really fancy (with 8 bits per channel, as almost all displays are, 64x or 256x are the mathematical "peak" where additional samples don't affect visual output anymore (less than 1 bit). In practice, 36x and higher rarely yield improvement (though that doesn't stop people from cranking AA up to the max in QuartzCrystal all the time, and then complaining that it doesn't render successfully anymore ;)

In OpenGL Driver Monitor, under Renderer Info, you can see the max samples -- does anyone have something higher than 4x? If so, we could do some testing to see if QC actually sets it to the highest supported, or if it just sets it to 4 as the default.

photonal's picture
Re: Iterated OpenCL Meshes

wow very cool clip!! just needs an audio track to go with it ;-)

toneburst's picture
Re: Iterated OpenCL Meshes

photonal wrote:
wow very cool clip!! just needs an audio track to go with it ;-)

I was thinking that myself. It would have to be a short track though, or I'd never get through it before it crashed again ;)

a|x

gtoledo3's picture
Re: Iterated OpenCL Meshes

Yeah... I was looking at my crash count w/ the last build of SL (since I'm going to update soon). 87! Yay.

usefuldesign.au's picture
Re: Iterated OpenCL Meshes

gtoledo3 wrote:
87! Yay.
In australia, 87 is known as the devils number in Cricket, a game you probably know little about. Batsmen are said to be at a high risk of getting out on that number (statistically unproved). !(Beware 87)

photonal's picture
was going to suggest

was going to suggest something - but just noticed your music site ... ... ... and now downloading 'we are america' :)