fix(esp_lvgl_port): Add support for LVGL 9.2 and above

[tore-espressif]

ESP-BSP Pull Request checklist

Note: For new BSPs create a PR with this link.

• Update benchmark project to use SIMD!
• CI passing

Change description

LVGL has changed their blending API here lvgl/lvgl#6688 , now it seems that it is considered private.
Moving forward, we plan to move the blending code to LVGL repo

[github-actions]

Test Results

0 tests   0 ✅  0s ⏱️
0 suites  0 💤
0 files    0 ❌

Results for commit 8c8f584.

♻️ This comment has been updated with latest results.

[peter-marcisovsky]

I can confirm that on this branch, the SIMD source code present in esp-bsp compiles successfully on LVGL >= 9.1.x

This solution solves both problems with SIMD src code support on LVGL 9.1.x and LVGL 9.2.x

1. changing structure names (_lv_draw_sw_blend_fill_dsc_t vs lv_draw_sw_blend_fill_dsc_t)

2. LV_DRAW_SW_ASM_CUSTOM_INCLUDE (Macro which we use to include our header file esp_lvgl_port_lv_blend.h with assembly function prototypes) being present in multiple places around LVGL (lv_draw_sw.c for example) and this .c file is missing include of lv_draw_sw_blend_private.h, causing compilation to fail

[tore-espressif]

@espzav PTAL :)

[github-actions]

No description provided.

[kisvegabor]

cc @uLipe

Looks great and very interesting! Two questions:

1. The benchmark was committed by @GitHub-Action. Do you run the benchmark with CI somehow?
2. What does the subscript numbers (e.g. (+17)) mean?

[espzav]

cc @uLipe

Looks great and very interesting! Two questions:

1. The benchmark was committed by @GitHub-Action. Do you run the benchmark with CI somehow?
2. What does the subscript numbers (e.g. (+17)) mean?

Hi @kisvegabor,

1. We have a runner for all boards (called BSP Wall Runner). It is for testing and we can run the benchmark example there (when PR is labeled - Run benchmark). Here is pytest for it: https://github.com/espressif/esp-bsp/blob/master/examples/display_lvgl_benchmark/pytest_display_lvgl_benchmark.py
2. The numbers are difference between master. In every push to master is benchmark run again and saved latest values to releases (https://github.com/espressif/esp-bsp/releases/tag/benchmark-latest). Each PR can compare the benchmark values with master.

[kisvegabor]

We have a runner for all boards (called BSP Wall Runner).

Wao, it's really impressive. Can you share a photo of it publicly?

The numbers are difference between master.

I see! It looks pretty good then for P4

I believe the "Rotated ARGB images" scene is this slow because it completely messes up the cache due to the "random" access of pixel in the original image. I haven't tested it, but moving the image to the internal RAM should make a huge difference.

However I don't know why the widgets demo is this slow. Only 16 FPS. Do you use LV_OS_FREERTOS LV_DRAW_SW_DRAW_UNIT_CNT 1 or 2? If not can you give it shot with these settings as well?

[espzav]

Wao, it's really impressive. Can you share a photo of it publicly?

Here is the photo

[kisvegabor]

This is really fantastic! We should have something similar for at least a few vendors and boards.