Odd CI Glitch

cybero's picture

Probably just a glitch in the Apple provided code, but the following attached example doesn't quite work as I'd been hoping to make it do.

I've deliberately stripped it down to more easily evidence the offending glitch, a dark line sporadically appearing. I don't think it is down to Particle System, was seeking a nice way to configure a multiple config for the patch, but am now vexed by this imperfection.

About the only type of simple workaround to this is to run add or over to get some kind of layered action.

coreimagecodeglitch

The glitch eludes ordinary screen grab , unlike outputting an image sequence of some Quartz Crystal rendering.

As far as I can make out , this is stock code from Apple, apologies to anyone else on the forum if this got into my Studies folder by any other means.

PreviewAttachmentSize
ciglitch.png
ciglitch.png17.85 KB
ciglitch.qtz15.57 KB

toneburst's picture
Re: Odd CI Glitch

That's quite a nice composition...

I'm getting the same glitch here (MacBook Pro 3.06GHz dual-core, currently running the GeForce 9400M GPU).

My guess is one of the GLSL functions used (Core Image filters use a subset of GLSL, and many of the same builtin functions), is returning an undefined value, perhaps when one of its inputs is 0, or negative.

Don't know if that helps fix the problem though.

a|x

.lov.'s picture
Re: Odd CI Glitch

nice comp!

Works fine with my MacBook (GMA950), no glitches!

psonice's picture
Re: Odd CI Glitch

Same here on ATI 2400. No glitches seen.

usefuldesign.au's picture
Re: Odd CI Glitch

Same system as one of your machines I think, dual G5 FX GeForce 5200 and yep glitch (intermittently like 20% of the time). Only 5 fps too.

cybero's picture
Re: Odd CI Glitch

Thanks for your comments.

Quote:

Same system as one of your machines I think, dual G5 FX GeForce 5200 and yep glitch (intermittently like 20% of the time). Only 5 fps too.

If I have no other composition rendering , then I get a good 60 fps full screen on my iMac and a good [•~] 5 fps on my dual G5 GeForce FX 5200 and no more than that.

That is by no means a CI Filter only performance issue, also hits on most other compositions. Also prompts me to get some screen size and chipset conditioned setups for some of my compositions. Still think it is worthwhile composing for earlier platforms, but some of my compositions might well be getting badged as best on Intel Mac .

As most of my development to date has been on PPC, my back catalogue, much unrevealed or unpublished will still stand me in good stead, might even be able to purpose PPC and Intel variants, or even chipset variants, but think that kind of refinement should be contained within the compositions own conditional set.

Quote:

My guess is one of the GLSL functions used (Core Image filters use a subset of GLSL, and many of the same builtin functions), is returning an undefined value, perhaps when one of its inputs is 0, or negative.

Is there some utility that my lack of CoreImage Filter work to date has left me unaware of available that can emulate, test, dry run or whatever, the graphics code? I guess in one sense, any CIFilter application, like Fun House, can provide a WYSIWYG type of feedback, I was looking for debugging.

(Looking to learn some more GLSL - CoreImage dialect :-)]

On a quick scan of the feedback, it looks like it is an nVidia specific issue, all Intel and ATI renderers prove good for the purpose.

Will recommence experimenting with CI Filter code a little later.

The actual composition that provoked this error was one in which I was playing around with Particle System / Interpolation Setups [Modulus|Power,..etc].

PreviewAttachmentSize
12s.qtz21.72 KB

psonice's picture
Re: Odd CI Glitch

I don't think it's PPC related. The nvidia 5200 cards are famous for being really slow for any kind of shaders. Somehow nvidia got that whole 5xxx series horribly wrong, they were all very slow, and the 5200 is the lowest end card of the series :( Most likely a 6xxx card (or even an ATI 9xxx from the same era) would run it much better. Apart from that, GPUs have come so far in the last few years that if you design for anything recent it's hard to get it to run well on older chips.

Btw, remember to submit this comp in a bug report to apple, if you're lucky it will get fixed :)

cybero's picture
Re: Odd CI Glitch

Good points - especially submit a bug report.

gtoledo3's picture
Re: Odd CI Glitch

I had thought that there weren't glitches for me, but when I scanned frame per frame I did see them two or three times (X3100) in a 60 second render. I really thought it hadn't glitched. I definitely don't see it in normal runtime, only in render... extremely fast, and not noticeable unless I scrub through the timeline.

I would be curious what happened if you reigned in the values on that interpolate so that you weren't doing a full 360 either way.

cybero's picture
Re: Odd CI Glitch

Yeah, could limit down between 0 to 360 , just done it, doesn't work out the kink unfortunately, it really is GLSL shader / Core Image related, back to more background reading.