Ah, that kind of makes sense. I stick with what I said in my last email though- if this kind of problem is inevitably going to happen with certain setups, and the Quartz Composer Editor allows you to setup a system where it does, and there isn't any kind of warning that it might, that's a problem. Maybe it's not a bug, though... Having said that, actually explaining how the problem might occur in a succinct way might present it's own challenge....
a|x --- On Wed, 9/2/11, Christopher Wright <christopher_wri...@apple.com> wrote: > From: Christopher Wright <christopher_wri...@apple.com> > Subject: Re: Image Feedback Memory Leaks > To: "Chris Wood" <pson...@gmail.com> > Cc: "Alex Drinkwater" <the_vo...@yahoo.co.uk>, "quartzcomposer-dev list list" > <quartzcomposer-dev@lists.apple.com> > Date: Wednesday, 9 February, 2011, 21:59 > > Here's the same thing with GLSL > instead of CI patches. No leak, still using the feedback > patch. No idea why this works and CI doesn't, surely the CI > version at some level gets translated into pretty much the > exact same thing? > > Not quite: CI doesn't composite results normally -- > CI Images are best thought of as "recipes"; filters to > apply to sources. If you have an image, and you apply > effect A to it, there are 2 images (the source, and the > source with A applied). If you then apply A to the > result, you have the original source image, the image with A > applied (as an intermediate), and the image with A applied > applied again. The filter itself probably does boil > down to comparable GLSL, but the source and intermediate > materials are never "flattened" (old intermediates aren't > discarded). CI Accumulator is the one instance where > "flattening" actually occurs; otherwise, it needs all > previous inputs to calculate the output. As you can > imagine, this piles up rather quickly. > > The shader version uses Render In Image, which is little > more than a fancy FBO. It doesn't need to keep around > intermediate results beyond the previous frame used as > input. This means it uses less memory, but also means > that it's somewhat less accurate with certain operations. > > -- > Christopher Wright > christopher_wri...@apple.com > > > > _______________________________________________ Do not post admin requests to the list. They will be ignored. Quartzcomposer-dev mailing list (Quartzcomposer-dev@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/quartzcomposer-dev/archive%40mail-archive.com This email sent to arch...@mail-archive.com
