Getting following error while executing the qnn-net-run using custom operator for Conv2d layer.
1.3ms [ INFO ] Loading model shared library ([model].so)
1.5ms [ INFO ] qnn-sample-app build version: v2.8.0.230223123141_52150-auto
1.5ms [ INFO ] Backend build version: v2.8.0.230223123141_52150-auto
1.7ms [ INFO ] Initializing logging in the backend. Callback: [0x40ab00], Log Level: [3]
1.7ms [ INFO ] Initialize Backend Returned Status = 0
1.7ms [ ERROR ] Device property not supported or not known to backend
[ ERROR ] QnnModel::addNode() validating node sequential_conv2d_Conv2D failed.
[ ERROR ] frozen_graph.addNode(QNN_OPCONFIG_VERSION_1, "sequential_conv2d_Conv2D", "ConvOp", "Conv2D", params_sequential_conv2d_Conv2D, 4, inputs_sequential_conv2d_Conv2D, 1, outputs_sequential_conv2d_Conv2D, 1 ) expected MODEL_NO_ERROR, got MODEL_GRAPH_ERROR
2.1ms [ ERROR ] Failed in composeGraphs()
2.1ms [ ERROR ] Graph Prepare failure
QNN SDK 2.8.0 is used
Need clarification on from where this error is generating? Is it model issue or xml file issue or the CPP implementation of custom operator?
Dear devleloper,
You could take reference as the UDO example code on QNN SDK.
BR.
Wei
In QNN SDK example there is no cpp implementation for the operators with Weights, There is only implementation of Relu operators in SDK.
Needed cpp implementation of operators with weights like Conv2D