perf: LVGL 8.3

[gagarinlg]

Summary of changes: switch to LVGL v8.3 which contains a number of bug fixes and faster the DMA2D usage has been improved

[pfeerick]

The tests for this failed as there has been a fractional colour shift (and I do mean fractional, I only noticed it when zooming in on the text - and text only, but the colour picker can see/quantify it)... any idea where that came from, and is it anything to be worried about?

Original:

Failed:

Original:

Failed:

[gagarinlg]

They did some changes to the font rendering with 8.3

[pfeerick]

Fair enough, I'll update the tests and continue on then :)

[pfeerick]

This looks mostly good with two minor issues

• The standalone lua scripts screen seems like it's either one pixel short, or one pixel higher on the 480x272 screen (but seems ok on the 320x480) - i.e. there is a single row of white at the bottom of the screen that wasn't there on a prior nightly on the TX16S

• default splash screen on the NV14/EL18 will need some TLC as the "blank" space above and below is now filled with in-memory garbage - I guess a screen clear or fill before showing the logo?

[richardclli]

For the splash screen, see if this can resolve the problem. #2764
I do not know how to work out the effect in #638, but it can replace the current splash screen with a dynamic background.

[pfeerick]

Yup, that'll do for now... (just tried it cherry-picked into this PR) so you're fine with that being merged even though you still have it marked as draft? 😆

[richardclli]

Yup, that'll do for now... (just tried it cherry-picked into this PR) so you're fine with that being merged even though you still have it marked as draft? 😆

I change the status of the PR, let's use it first.

[gagarinlg]

Moved to 2.10, because it is waiting for libopenui changes

[pfeerick]

And more importantly, it looks like there are still some kinks to be worked out.

[raphaelcoeffic]

@gagarinlg @philmoz let's tackle this in September :-) It seems the 8.3.x series is getting mature now (currently at v8.3.9).

[pfeerick]

I think this issue

The standalone lua scripts screen seems like it's either one pixel short, or one pixel higher on the 480x272 screen (but seems ok on the 320x480) - i.e. there is a single row of white at the bottom of the screen that wasn't there on a prior nightly on the TX16S

is still present, and it could actually be out one row at the bottom, and on the right side (making it one pixel down and one pixel in, since the screen is flipped?). It's most noticeable for me with Offer's Model Flight Log viewer (https://github.com/offer-shmuely/edgetx-x10-scripts/tree/main/SCRIPTS/TOOLS/LogViewer), as I get a white line on at the bottom and right side when the splash screen is being shown, but I do get it other times also.

[philmoz]

I think this issue

The standalone lua scripts screen seems like it's either one pixel short, or one pixel higher on the 480x272 screen (but seems ok on the 320x480) - i.e. there is a single row of white at the bottom of the screen that wasn't there on a prior nightly on the TX16S

is still present, and it could actually be out one row at the bottom, and on the right side (making it one pixel down and one pixel in, since the screen is flipped?). It's most noticeable for me with Offer's Model Flight Log viewer (https://github.com/offer-shmuely/edgetx-x10-scripts/tree/main/SCRIPTS/TOOLS/LogViewer), as I get a white line on at the bottom and right side when the splash screen is being shown, but I do get it other times also.

Is it reproducible in the simulator?

[philmoz]

@pfeerick If you mean like the image below then this happens in the simulator on both 2.8.4 and 2.9.1 as well.

The problem is the image itself has these lines at the bottom and right edge.

[pfeerick]

Thanks for checking that... I hadn't tried on simulator, but now that I had what I thought was an easily reproducible test case I tried it on nightly and that was the same. I also just noticed the ELRS Lua has a thin border on the left right and bottom sides, and that is visible on this branch, so whatever caused this before is clearly gone. Now it's just a matter of any final changes to the LVGL 8.3 branch.

[philmoz]

There is something going wrong though.
I get this line at the bottom on the EdgeTX 3.0 UI demo script on my TX16S; but not in the simulator.

Does not happen with current main.

Unfortunately as it's distributed pre-compiled it's not possible to see how the image is being displayed and debug the code.

[pfeerick]

I've seen something exactly like that before, so that could have been what was happening before (it was a few months ago 🤭), or I've seen that same glitch elsewhere. 🤷

[philmoz]

I did some investigation into the white bottom line issue with LVGL 8.3.

It seems to be some timing issue or conflict with the DMA used to copy the standalone script bitmap and the DMA used to rotate the screen.

The rotate screen code does a line at a time starting at the bottom row of the source and working up. If I change the direction so it works from the top row of the source and works down the problem goes away.

[pfeerick]

Looks like the structure of lv_style_t changed in 8.3... so this is not playing nicely with #4180.

[philmoz]

Looks like the structure of lv_style_t changed in 8.3... so this is not playing nicely with #4180.

Should be fixed now.

[pfeerick]

TX16S and EL18 both seem ok now with this PR... no immediately visible issues after loading several themes, trawling through some images, running some widgets and standalone Lua like Offer's graphing log viewer...

[robustini]

Horus X10S Express: when I change the points of a curve the selection has tremendous lag.
With LVGL 8.2 it was fine.