Improve detect events skips

[automech-rb]

Hi,

Thanks for this awesome C++ implementation.
I am using the GPIO::add_event_detect(channel, GPIO::BOTH, callback_fn) feature to detect edge in Jetson AGX Xavier digital input. Environment is JP5.0.2 with docker.

I have a rotary encoder whose digital output change every 1 milliseconds from high to low (and vice versa).

The problem is when CPU load increases on Jetson to 50% or more, the add_event_detect is not able to pick up all the encoder output change. It don't detect around 30% of the edge (less callbacks) for same encoder RPM. I understand that this is not a RTOS based system, but do you have any recommendations to improve the GPIO callback performance?

thanks

[automech-rb]

To add, this error behaviour is easy to simulate when comparing output if 15W and MAXN mode.

For now to mitigate this issue, I am using different thread and looping at >1000hz to check digital input.

Also, #77 seems to be similar issue but for 20KHz instead of 1KHz for mine!

[pjueon]

Thanks for the report.

@ShimmyShaman
Do you have any idea about this issue?

[ShimmyShaman]

Distracted by other life things, but I have been looking into it. Believe there are fixes that can be made, but have to test/explore them.

[pjueon]

@automech-rb
Does this issue also occur when CPU load is low?

[automech-rb]

From my testing, this issue only occurs with high CPU usage (over 50%).
To add more details:

• I am using two add_event_detect for two encoders running inside docker container on JP5.0.2
• Similar error behavior is observed when compute performance is limited by changing mode to 15W (+ more erratic callback sometimes low sometimes high in same timespan for constant encoder RPM)