Drawing latency compared to xochitl

[Evidlo]

I've been playing with the demo and I noticed that the pen latency is a bit higher than in xochitl. Any idea what the cause is or if it can be reduced?

[SafariMonkey]

I'm sorry that this was somewhat forgotten, I wrote a long reply when it was originally posted and accidentally closed the tab and lost it. I've been meaning to rewrite it but have been distracted by other things. I'll try to give a shorter summary of those points. I'd appreciate @canselcik's thoughts too, if/when he has time.

We haven't been able to reliably test the latency, due to lack of high quality high speed recording equipment. Subjectively, it seems to me to be fairly competitive, but I do think it's a little slower.

The version of the libremarkable demo on the master branch (currently unreleased I believe) uses a more efficient drawing implementation and a Bezier based drawing method, giving only an additional 0.5 samples of average latency over linear interpolation. (Given the expected report rates of 200+ Hz (untested), that would be less than 3ms additional latency. As such, I doubt that's the issue.) If you haven't tried this version, I'd be interested to know if you think it's faster, and how it compares to the xochitl version.

For other possible factors, there's this from an article about the rM:

He set himself the goal of removing everything that gave the system inertia, from the operating system to the hardware and software. In addition, we have worked a lot with prediction. The software calculates the position of the pen in the future and draws the line where it comes, says communication manager at Remarkable, Henrik Faller.

(via Google Translate)

That seems to be indicating that xochitl is actually using predictive drawing, which may be the case but I haven't managed to fool it yet (e.g. by drawing quickly up to the edge of a ruler) so I'm not totally sure that it's implemented in the available version.

[LinusCDE]

The difference should be negligible and most people perceive the latency to be the same. Apart from a minimal function diff, the difference will likely be in how the drawing is done (naive, bezier, ...).

I doubt that there is an actual low level difference and latency on the rM 2 should currently be objectively higher due to the need to send calls via IPC (but most likely negligible as well).

We haven't been able to reliably test the latency, due to lack of high quality high speed recording equipment. Subjectively, it seems to me to be fairly competitive, but I do think it's a little slower.

If anyone has such equipment, comparing the demo (and basic_draw example) to xochitl with it would be really valuable on either model. Apart from that, I don't think there is much we can do to fix this.