|
|
Structure ProcessorTowards a processor equivalent of the iterator renderer. A structure pipe + grep. A hint of what spreads could do. Take your metaphor of choice! A macro patch that takes in a structure and then passes each structure member to its subpatches and outputs the modified structure. Would need an 'iterator variables' equivalent to handle the feeding in and out of the subpatch, and the subpatch would need to know the key or index of the structure item its currently processing. Lets say I have a structure of images and positioning metadata. With this, I could add a filter to those images and leave the metadata alone. Or a bunch of 3D geometry, GL lines etc, now you can manipulate each coordinate. For me, this is a wall I've hit as its beyond the scope of the official api. Toby
|
I've been thinking about some stuff like this (basically, mid-graph loops for sets of data). The problem currently with this is that the Iterator patch is red, so it can't be used as such. a non-red iterator, however, would probably do the trick for this, wouldn't it?
structure in -> iterate through each structure element doing whatever you please -> structure out.
second that...
alx
Quartz Composer Blog: http://machinesdontcare.wordpress.com
Music Site: http://www.toneburst.net
wouldn't it be possible to this kind of thing with JavaScript? Admittedly, this would require a lot of coding, mind...
alx
Quartz Composer Blog: http://machinesdontcare.wordpress.com
Music Site: http://www.toneburst.net
I have tried to iterate maths... it seems to work with a spooky patch loop inside an iterator... However the structure assembly need to be improve a lot ! I don't think that a lot of spooky channels would be very efficient but it is just a way to explore...:)
I've not got spooky send installed here (or time to test it really), but do you not need to rebuild the structure with a queue patch (or structure tools)?
and more generally, an enhanced version of the structure tools, that's in the air for quite a long time now ( i guess)
What kinds of stuff would y'all like to see in an enhanced structure tools patch?
as a general statement, maybe it is time to implement single dimension c-arrays as a data type that can be passed between kineme structure tools and kineme gl tools? we could call the data type 'stream', and i think it would really clarify whats going on when we're trying to patch this kind of world together. structures are for structured information, streams are for streams of the same data type, typically coordinates.
...and we probably shouldn't call them spreads, as a) streams is more accurate for our purposes here, and b) the might of spreads lies more with the concept of vvvv nodes being spreadable by design (ie they iterate themselves correctly and efficiently) rather than just having the plain data-type.
would be a single dimension plain c array (ie as fast as a computer can go), with the x,y,z coordinate information encoded as... x,y,z,x,y,z,...,x,y,z. or r,g,b,a...r,g,b,a or etc.
that would come in handy when handling serial streams from the arduino as well!
• possibility to pass the structure thru 2D/3D/4Dnoise...
• Structure manager : to edit vertex on the fly, with basic ACCII import/export. ( a patch with a table as a custom interface to display : index and vectors )
•Structure blend : input 2 structures with same lenght and same array number and apply blend mode (mix, add, divide, multiply,max, min..etc) then output result.
•"Convert to object" "patch : to plug structures to Kineme 3D Object render...in this case the structure could fly thru the existing Kineme 3D Patches...:)
•1D structure generator : generate an array with custom lenght (with decrease, increase, sin, cos Arrays ) and minimum/maximum amplitude.
any tool that can remove nan. entries will do. (or any specific entry) then any possibility to merge, sort, find, reorder, multiply, divide, duplicate... would be a great improvement anyway. and some pizza
Ok, huge turn out for this one I guess ;P
I've smallified the structure maker (working on finishing touches for removing the skip empty inputs port via checkbox). Next up, I think I can make the 1D structure maker pretty simply, as well as the structure merger. Noising is also easy, except for strange types (string noise? image noise? those could work, but are expensive, so I'll probably skip them for now) -- 1D noise only I suppose.
blend should be renamed "structure operation" or "structure op", because it's doing a set of operations between structures. not too difficult, though again we hit the "strange type" problem mentioned above.
I feel like the green iterator, combined with a few of the above, would make solving many issues possible.
streams suffer from the typical problem of having no built-in support. The muxing idea wouldn't be difficult, but I wonder if it's too specific (a more general "raw data" abstraction could be handy, but I don't know what that abstraction is yet... not given it much thought) -- this is slightly dangerous territory because introducing half-way solutions (like the last GLTools "spreads") isn't a real full solution, so it's bound to get passed up. That's a lot of extra work to recode everything using the half-way method.
and pizza. I think that would solve basically all the problems we're facing ;)
Do tou think that a "structure float array truncate" patch could optimize arrays flow ? (something like the string truncate but to truncate floats after the comma)
floats are a fixed size (32bits), so passing them around with or without a fractional part doesn't affect their memory foot print. Coincidentally, integers are also 32bits, so no memory is saved by converting floats to ints.
Converting floats to ints takes a few instructions per item (a nanosecond or so), truncating them takes a few more instruction (still only a nanosecond or so), but leaving them alone takes no time :)
They'd flow just the same either way, but the extra processing would actually slow things down (probably imperceptibly though, unless the arrays were gigantic)
so floating point precision is not "a luxe " ;) especially when 2D coordinates are from 0.0 to 1.0 !
thanks, this is really appreciated. I'm also personally crossing the need of a way to remove one specific entry from the structure (or even a whole substructure), so a delete feature of any fashion is super+