Proper Memory Allocator for Android QNN EP on GPU?

[brady-cherish]

What is the proper way to allocate memory/buffers (Ort::MemoryInfo and/or Ort::Allocator) for inference on Adreno GPU when using C++ APIs and the QNN EP?

I've had behavior where I create input tensors with varying data, however the results of subsequent inferences results in the same, repeated output.

• I've validated the input values are truly instantiated and not an empty buffer.
• I believe that memory might not be properly allocated to run on the QNN EP on GPU or that the input/output tensors are not properly copied to or from each.

Any help or pointers to appropriate resources would be greatly appreciated.