How to register my own provider's MemcpyToHost/MemcpyFromHost Op

[lewisword]

problem describe:
I follow the instruction of AddingExecutionProvider, and successfully register my own provider and some simple op(Add .etc). But when I run the onnx model which contains op both from CPU and my own provider, onnxruntime sends me one error that my own provider doesn't have MemcpyToHost/MemcpyFromHost Op.So how to?

System information:
onnxruntimev1.5.1

Additional context:

• I try to find the definition of MemcpyToHost/MemcpyFromHost in cuda, but fail.
• I try to register MemcpyToHost/MemcpyFromHost just like other common op(for example, Add), but fail

[galdav117]

I am encountering the exact same problem, could you share how you solved it?

Thanks!