LVGL performance dropped when migrating from zephyr v2.7 to v4.0/4.1.

[glory-man]

I use imxrt1060-based board with connected 768x1024 display. After migration to new Zephyr versions which uses v8(v9) LVGL instead of v7.6 UI performance dropped 2-3 times. Structure of UI-project fundamentally has not changed - where I could I refactored, where methods changed I replaced them with suitable ones. System configuration - for example CPU freq - the same as in zephyr 2.7. But the time required to render images has increased significantly.

some info at LVGL forum.

Maybe i missed something. How can i improve performance?

UPD: All data - OS heap, LVGL heap, files content - are stored in external SDRAM, accessing via SEMC.

[JairGudino]

Hi!

After reading your information, I don’t have a specific solution, but I do have some suggestions to help improve performance:

Double Frame Buffer: LVGL supports using a double frame buffer to reduce processing time. Make sure this feature is enabled.
Frame Buffer Location: You can move the frame buffer to SRAM for better performance. This application note can serve as a reference. https://www.nxp.com/docs/en/application-note/AN14597.pdf
Main Stack Size: Increasing the main stack size might help in some cases, although your issue may not be directly related to it.
Frame Buffer Rotation: If you're rotating the frame buffer to change the display orientation and the PXP is disabled, processing time could increase significantly.
I hope these suggestions help!

[glory-man]

Double Frame Buffer: LVGL supports using a double frame buffer to reduce processing time.

Yes. I use screen rotation. And current imxrt-elcdif driver supports rotation only if PXP enabled & full frame buffer should be roatated.
So my configuration

CONFIG_LV_Z_VDB_SIZE=100
CONFIG_LV_Z_FULL_REFRESH=y

1024x768x16bit color requires 1.5Mb of ram for frame buffer with this config. So I can't move frame buffer into SRAM because of limited SRAM space.

I tried to enable CONFIG_LV_Z_DOUBLE_VDB=y without significant changes in performance.

[glory-man]

My application uses bmp-images as background and some png-images as screen elements. And it seems like increased images read-time. If I switch screen2->screen3 full switch time increased to 500-600ms (from 150-250 ms in z2.7), but next time screen2->screen3 changed switch time 250-350 ms, as far as i understand, due to the use of images caching - png opened, decoded and stored in cache as decoded. But full read time of bmps increased to 50-70ms