You, sir, are a genius! That solved my problem beautifully. And it seems that only the producer of the buffer needs 32 bit Render In Image - consumers can be lowly 8 bit workers and still access the information.
Thank you a thousand times... Patrick On May 28, 2014, at 3:11 PM, Christopher Wright <[email protected]> wrote: > You are aware that OpenGL clamps a lot of values internally, correct? You'll > need to use "32 bit" Render in Image patches to get negative values to work I > believe (you may still be at the whim of what various vendors do). Most GL > texture formats clamp, so this can be tricky to deal with :( The implicit > clamp can be super annoying for GPGPU type workflows, as you've discovered. > > Anyway, here's a quick demo for negatives (I didn't go beyond +-1 though, but > it probably functions there too). Does it more or less do what you're > looking for (you should see a red spinning rectangle where the negative > values are)? > <negative.qtz> > > You'll probably be stuck working explicitly in Alpha = 1 domain, since QC > likes premultiplication a lot, and premultiplication will destroy fidelity at > best, and at worst will do the wrong thing entirely. > > Also note that "32 bit" here means 32 bits per component, so each pixel will > consume 16 bytes of vram (4 32-bit floats). This can be rather taxing for > some GPUs, as the textures are much larger than more pedestrian > 8-bit-per-component images, and might exhaust vram if your images are > sufficiently large. Fillrate will also be adjusted accordingly (~25% of the > normal pixels per second). > > On May 28, 2014, at 2:47 PM, Patrick Sheffield <[email protected]> > wrote: > >> >> I've been having a devil of a time with this in Quartz Composer. >> >> I wanted to analyze an image, store the information in a buffer, then use >> that buffer for subsequent iterative processing of the image (why I need to >> store the information first). >> >> The problem is that the analysis can produce positive and negative numbers >> (and potentially under -1 or over 1) and Quartz Composer won't leave the >> values alone EVEN if I calculate them using a GLSL program inside a Render >> In Image with the flag set to Disable Color Correction. I think it simply >> clips off the negative values. >> >> I tried compressing the range by adding 1 and dividing by 2, but that is >> less than effective. I also tried the secret preference of turning off >> correct Core Image. That had no effect, plus I can't guarantee it'll be >> turned off on another system. >> >> Is there ANY way to store a buffer of information and access it later >> without QC diddling with it when it enters a pipe? >> >> Thanks in advance, >> >> Patrick >> _______________________________________________ >> Do not post admin requests to the list. They will be ignored. >> Quartzcomposer-dev mailing list ([email protected]) >> Help/Unsubscribe/Update your Subscription: >> https://lists.apple.com/mailman/options/quartzcomposer-dev/christopher_wright%40apple.com >> >> This email sent to [email protected] > > -- > Christopher Wright > [email protected] > _______________________________________________ Do not post admin requests to the list. They will be ignored. Quartzcomposer-dev mailing list ([email protected]) Help/Unsubscribe/Update your Subscription: https://lists.apple.com/mailman/options/quartzcomposer-dev/archive%40mail-archive.com This email sent to [email protected]
