Mobile performance

[davepagurek]

Topic

As p5 sees increased adoption for mobile AR/VR, what can we do to make sure things run more smoothly? Here are a few points to discuss:

• After Stroke cap+join support for WebGL #5802, stroke rendering is a bit heavier than it used to be. Is this a bottleneck? Should we add a NONE stroke cap/join type so that applications targeting mobile can opt out of the extra geometry of higher fidelity strokes?
• After Modify the shader so that the line can be colored per vertex #5915, per-stroke colors are used to allow interpolation in the same way fill interpolation works. Is this a bottleneck? Should we offer methods to disable all per-vertex colors for the sake of performance?
• Drawing the same shape repeatedly would be more efficient if it is cached on the GPU as a p5.Geometry. Is this an optimization that most sketches would benefit from? If so, should we provide an easier way to create these? (The ability to construct custom p5.Geometry using existing modeling commands. #5393)

In general, it seems like we should do some profiles to get some data on what parts are slowest on mobile in order to decide what approach would help.

[theVJagrawal]

Can I work on this issue? I am pretty new to open source stuff

[davepagurek]

Thanks for your interest @theVJagrawal! Probably the first step for this would be to find some WebGL sketches that run OK on desktop but not on mobile, and see if you can run some benchmarks (e.g. through Chrome remote debugging) and share some profiler results to see what methods are taking up most of the time each frame.

[davepagurek]

I did an experiment today seeing if tessellation in WebGL mode could be improved by using a different tessellation library. Mapbox's earcut library claims to be a bit faster and I made a sketch testing it out https://editor.p5js.org/davepagurek/sketches/gmd_WPXX9 but it's not all that much faster (like ~5ps better?) while handling self intersections less cleanly.

So probably rather than optimizing immediate mode more, the better thing to do is make it easier for people to use retained mode (i.e. #5393).