Can I use TensorIterator when I writing pytorch C++/CUDA extension?

I find there are many TensorIterator ops in pytorch source code. But I can not use TensorIterator when I writing C++/CUDA extension? I think TensorIterator is very important. Without TensorIterator, the program will have unpredictable bugs. How do I solve this problem?

Have a look at this doc to for an example.

When I writing C++/CUDA extension, I can not find this head file

#include <ATen/native/Lerp.h>
#include <ATen/native/TensorIterator.h>
#include <ATen/native/cpu/Loops.h>

You wouldn’t need Lerp for your extension, would you?
Are you able to fine the other headers?

Unfortunately, I can not find the other headers. I install pytorch using pip3. I only find this files.

At same time, I can not find this header file in libtorch. I can only find this files as follows