Try changing GUI_BYTESPERPING back to its original value. -Jonathan
On Thursday, January 7, 2016 12:21 PM, Ivica Ico Bukvic <ico@vt.edu> wrote:
I would test the same patch you built. To tax tkpath and tk in particular you can spread toggles to the four corners of the screen since redrawing area calculation on those toolkits is fairly inefficient. CPU spec and CPU overhead data would also help figure out where the bottleneck lies. e.g. my experience is from a low-powered AMD APU (essentially AMD's Atom counterpart).
On 1/7/2016 11:52 AM, Jonathan Wilkes wrote:
Do you have a simple patch to test? Not sure exactly what's going on, but I can think of three relevant areas: * node.js APIs are largely built to be asynchronous * V8 javascript engine does JIT compilation of hot code * Chromium goes out of its way to ensure that nothing blocks the renderer from doing its job Of course advertisers don't have to pay for the CPU they consume on your device, which mean all browser-based technology are CPU hogs. -Jonathan
On Thursday, January 7, 2016 11:19 AM, Ivica Ico Bukvic <ico@vt.edu> wrote:
Would you please test the same on pd-l2ork and/or vanilla? If there is a stark difference, this may suggest that node webkit does some kind of internal event pruning (which would be really sweet). Also, checking CPU overhead may help shed some light as to what is going on. Finally, what are the specs of your machine?
On 1/7/2016 2:20 AM, Jonathan Wilkes wrote:
In the GUI port I hooked up about 14 toggles to a [metro 1] and was getting
anywhere between 30 and 40 fps in the GUI. I could continue creating new objects in the same patch. Now if I connected each of those toggles to a [print] I'd probably freeze the GUI. But I'm not doing any optimization on console, nor limiting the size of the scrollback buffer. (I am, however, displaying a hit count beside duplicate messages which cuts down immensely on noise.) At least for now, if some common patch behavior is able to DOS the GUI I'd rather have it freeze up as an incentive to fix the underlying problem. -Jonathan
On Wednesday, January 6, 2016 1:07 PM, Ivica Bukvic <ico@vt.edu> wrote:
Create a metronome with less than 2 milliseconds clock and connect it to a bunch of toggles, turn it on, and enjoy interacting with a barely responsive gui. The same is achievable with a less dubious implementation where a graphical user interface simply has a lot of concurrently animated components. NB: depending on your computer specs the two millisecond threshold may have to be set lower. --
Ivica Ico Bukvic, D.M.A. Associate Professor Computer Music ICAT Senior Fellow Director -- DISIS, L2Ork Virginia Tech School of Performing Arts – 0141 Blacksburg, VA 24061 (540) 231-6139 ico@vt.edu www.performingarts.vt.edu disis.icat.vt.edu l2ork.icat.vt.edu ico.bukvic.net
On Jan 6, 2016 9:31 AM, "Jonathan Wilkes" jancsika@yahoo.com wrote:
Are there more than two people (matju and Miller) who understand the
_current_ design? -Jonathan
On Wednesday, January 6, 2016 12:00 AM, Ivica Ico Bukvic <ico@vt.edu> wrote:
The following is going somewhat OT but nonetheless based on this very interesting topic: application framerate is no more deterministic than the proposed idea. Just because monitor refreshes at 60Hz doesn't mean the app will do the same or consistently. Yet, pegging GUI updates at 60Hz (think speedlim-like behavior) despite all the loose ends will undoubtedly provide improved responsiveness over a GUI where one can send sub-millisecond change requests, most of which will never see the light of the day, despite eating up tons of CPU and in all likelihood bringing it down to its knees.
On 1/4/2016 9:32 PM, Jonathan Wilkes wrote:
Having an animation method allows GUI-side optimizations that aren't
possible when the animation is smeared across the socket/parser/eval'er. Determinism: GUI rendering isn't deterministic, at least not in the way Pd's scheduler is. For example, how could the GUI (tcl/tk or otherwise) even know that it missed a deadline?
I guess the simple API I'm toying with is "stretching the accordion" so to speak, and potentially showing the cracks more explicitly with a longer running animation than can currently be seen in Pd's GUI. But that can be remedied, either by recursively halving a single animation into smaller ones, or just giving up and using [line].
The greater benefit is that more elegant types of visual feedback become possible without being discarded out of hand due to their potential for audio interruption. -Jonathan
On Monday, January 4, 2016 5:20 PM, Ivica Ico Bukvic <ico@vt.edu> wrote:
On 1/3/2016 3:24 PM, Jonathan Wilkes wrote:
It's actually way more simplistic than that-- just an "animate" method that
wraps around whatever attribute is available to the drawing command. All I'll have is a ramp time, an optional delay time, and an optional easing curve. That will make it a bit like a [vline~] for GUI side animation. I'm using the web animations API because it makes things very simple, even to do complex things like animating path data. The idea is that this would open up some modest visualization opportunities that are otherwise too cpu intensive. For example, if you're animating using [line] the socket messages can quickly get in the way of the audio.
Will you at some point drown the CPU? Sure, but that is no different than a million of other ways of doing the same. OTOH, implementing the animation this way helps you ensure that your animation remains in sync with the audio, which to me seems much better gain than a potential CPU/socket overhead may be a shortcoming.
There could be some very cool ways of filtering gui messages, as well (short of getting rid of the socket-based communication in favor of a shared memory/multithreaded design). For instance, your animation object could be given the screen refresh rate and therefore it would not send out a message via a socket unless a desired frame-worth of time has transpired since the last message was sent. This would do wonders not just in terms of animation, but also the overall gui responsiveness, if implemented system-wide.
Best,
Ico
-Jonathan
On Sunday, January 3, 2016 1:08 PM, Ivica Ico Bukvic <ico@vt.edu> wrote:
I think it may make sense in addition to having a one-shot-independent animations that have no guarantee of staying in sync with the audio (e.g. these could be useful for mouse-over button animations) that your animation object can also receive a decimal value between its originator and destination, allowing for each keyframe to be a whole number. So, 0-1 would interpolate between the starting state and first keyframe, 1-2 between first and second keyframes, etc., and thus allow pd to use its timing mechanism to project changes in animation state via a line object, a counter or something similar. IIRC most (all?) HTML5-based animations can be triggered as independent events or can be given a specific percentage value. The one-shot object could interact with independent events, while the proposed object could interact with the latter.
That said, not knowing how you have imagined your animation object, it may be tricky to implement this as it would require object to keep track of all the keyframed events (assuming there are more than one). If you are thinking of having the animation object track only one single animation (e.g. something progressing from 30% to 90%), the same could still prove useful except in this case you would only allow for values between 0 and 1.
On 1/2/2016 1:12 PM, Jonathan Wilkes via Pd-list wrote:
Hi list, I'm playing with adding a simple animation api to data structure drawing commands. The parameters will be sent to the GUI, and the GUI will take care of the ramp, delay, etc. I'm thinking of just making it a simple "set it and forget it" api. That is, you send a message with your ramp and delay times to the GUI, and you just blindly trust that the GUI will make things happen in the right amount of time. The alternative I can think of is to have the GUI call back when an animation is finished, but that would encourage mixing the two clocks (i.e., GUI and Pd clock) in unpredictable ways. Does this simple approach seem like a reasonable design? The biggest problem would be that a long-running animation could skew. But in that case you could probably amortize the cost of sending more messages over the longer time period.
-Jonathan
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list