Hi, I am currently trying to run a DLA model compiled for MDLA5.3 on the G520 EVK.
I am using the Neuron SDK package for MT8189 provided in the “Neuron SDK” link. On the host side(Ubuntu 22.04), I used the ncc-tflite tool under the host directory to convert the model into .dla format using the following command:
./ncc-tflite --arch=mdla5.3,edma3.6 ./qat_quant.tflite -o ./qat_quant_g520.dla --relax-fp32 --opt=3
The conversion completes successfully.
However, when I run the test using neuronrt with the following command:
./neuronrt -m hw -a qat_quant_g520.dla -i bin_input/1.png_2828.bin -o output_g520.bin -c 10 -b 100
I encounter the error below:
WARNING: Cannot create device for APUSYS_2_0
WARNING: Found an unsupported target: APUSYS_2_0
WARNING: Fail to revise dla::CompiledResult
WARNING: Fail to preprocess dla::CompiledGraph
WARNING: Fail to preprocess dla::CompiledNetwork
ERROR: Cannot prepare execution.
ERROR: Successfully open network but cannot start execution.
ERROR: NeuronRuntime_loadNetworkFromFile fails: Cannot load network
Cannot load model from file qat_quant_g520.dla
Read dla fail
What I find strange is that if I push the mt8189 directory to the G520 via adb push and perform the same steps directly on the device, it works properly.
May I ask what could be the reason for this error occurring only on the host side?
Thanks!