How does torch.jit.tracer deal with precompiled functions?

My neural network includes a number of PyTorch layers, e.g. nn.Linear, nn.Conv2d, etc. and a number of precompiled functions, e.g. functions written in pure C/C++ or C/C++ with ATen data types.

When I run the tracer, I don’t see calls to these precompiled functions in the tracer.code. Should I expect to see them in the generated code?