Issue: Application Stuck When Running TFLite Model on NPU (A55) While M33 Core is Executing .ELF Bin

[jagadesh]

I'm currently working on a system where:

• Cortex-M33 core is running a "hello_world.elf" binary.

• Cortex-A55 core is responsible for running a TensorFlow Lite model using the NPU.

The problem I’m facing is that when the M33 core starts executing the "hello_world.elf" , the TFLite model on the A55 side gets stuck - the model not loaded , and inference stucks to execute.

Any suggestions or debugging tips would be greatly, Please give the reason behind it.

[Manuel_Salas]

Hello @jagadesh 

I hope you are doing very well.

This behavior is expected. As you can see on the ethosu_device.c driver, the rproc is loading the ETHOSU_FIRMWARE firmware to the Cortex M33.

Why?

This can be confirmed on the i.MX Machine Learning User's Guide, in figure 2:

Manuel_Salas_0-1744917787456.png

The Ethos-U runs in the Cortex M33 of i.MX93.

Manuel_Salas_1-1744917846396.png

That is the reason why you are having that issue.

I hope this can helps to you.

Best regards,

Salas.

[jagadesh]

root@ccimx93-dvk:~# cat /sys/class/remoteproc/remoteproc0/firmware
ethosu_firmware
root@ccimx93-dvk:~# [Apr17 15:35] imx-rproc remoteproc-cm33: can't change firmware while running[Apr17 15:35] imx-rproc remoteproc-cm33: can't change firmware while running

Please add suggestions.