Gabriel Wang

@kisvegabor When we want to see how it looks, we don't need a dedicated VISUAL mode but enable the low-level flushing, i.e. call `disp_enable_update()` before `lv_demo_benchmark()`.

@kisvegabor OK, the two combinations sound reasonable to me also. And please don't forget, this PR exposes a problem, which is why `2` instead of `1` get the best performance?

Thank you very much! @kisvegabor

@kisvegabor Have you disabled disp_flush? There should be significant differences between 1 and 2. And what kind of platform do you use? We observe differences on MCU platforms.

@kisvegabor I am using the MPS3 FPGA prototyping system, which provides an idea condition for testing: - Zero wait state RAM (ITCM and DTCM) - No cache While STM32F769 runs...

@kisvegabor I think I found the difference. In our previous testing, macro `LV_DISP_DEF_REFR_PERIOD` was configurable, and we found that 2 was better than 1. As discussed in the issue https://github.com/lvgl/lvgl/issues/3530,...

Yes, I am using the latest master. The conclusion is the same, 2 is significantly better than 1.

> Are there tests where you got >500 or >1000 FPS? Yes, Some scenes got > 1000 FPS. Similar to this https://github.com/lvgl/lv_port_an547_cm55_sim/blob/master/lvgl_benchmark.md

@kisvegabor I am using 320*240 RGB565

OK, I will have a try and let you know the result.