NPU models don't work on Linux Kernel 5.17.0

VIM3
#1

Which Khadas SBC do you use?

VIM3 Pro

Which system do you use? Android, Ubuntu, OOWOW or others?

Ubuntu

Which version of system do you use? Khadas official images, self built images, or others?

self build docker image on Ubuntu 20.04.4 LTS 5.17.0

Please describe your issue below:

demo script (yolov3-picture.py) worked fine on Ubuntu 20.04.4 LTS 5.16.0 and 5.13.0 but after we’ve moved it to Ubuntu 20.04.4 LTS 5.17.0 it’s stopped working
$ python3 yolov3-picture.py --model ./models/VIM3/yolov3.nb --library ./libs/libnn_yolov3.so --picture ./data/1080p.bmp

Post a console log of your issue below:

|—+ KSNN Version: v1.2 ±–|
Start init neural network …
[ 1] HAL user version 6.4.6.2.345497
[ 2] HAL kernel version 6.4.6.2.345497
E [vnn_CreateYolov3:187]Create graph fail.
E [vnn_CreateNeuralNetwork:208]CHECK PTR 208
Done.
Get input data …
Done.
Start inference …
Segmentation fault (core dumped)
$
``

#2

@Pavel1

  1. Please run sudo dmesg | grep Galcore to check you system npu driver version.

  2. The least KSNN is V1.3, please use the least KSNN code to test

I burn the 5.16-rc2 image into SD card, then upgrade it, success to upgrade NPU to 6.4.8.7.

#3

$> dmesg | grep Galcore
[ 3.603723] Galcore version 6.4.8.7.1.1.1

I have upgraded KSNN to V1.3 (it was v1.2) but it still doesn’t work:
$> python3 yolov3-picture.py --model yolov3.nb --library libnn_yolov3.so --picture 1080p.bmp
|—+ KSNN Version: v1.3 ±–|
Start init neural network …
[ 1] HAL user version 6.4.6.2.345497
[ 2] HAL kernel version 6.4.6.2.345497
E [vnn_CreateModel:196]Create graph fail.
E [vnn_CreateNeuralNetwork:208]CHECK PTR 208
Done.
Get input data …
Done.
Start inference …
Segmentation fault (core dumped)

#4

I have installed the new aml-npu_6.4.8.7.1.1.1_arm64.deb and the new models from KSNN repo - now it all works, thank you @Frank !

1 Like