ether ``cuda_sources`` is
            provided. Set it to ``True`` to force CUDA headers
            and libraries to be included.
        with_pytorch_error_handling: Determines whether pytorch error and
            warning macros are handled by pytorch instead of pybind. To do
            this, each function ``foo`` is called via an intermediary ``_safe_foo``
            function. This redirection might cause issues in obscure cases
            of cpp. This flag should be set to ``False`` when this redirect
            causes issues.

    Example:
        >>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_CPP_EXT)
        >>> from torch.utils.cpp_extension import load_inline
        >>> source = """
        at::Tensor sin_add(at::Tensor x, at::Tensor y) {
          return x.sin() + y.sin();
        }
        """
        >>> module = load_inline(name='inline_extension',
        ...                      cpp_sources=[source],
        ...                      functions=['sin_add'])

    .. note::
        By default, the Ninja backend uses #CPUS + 2 workers to build the
        extension. This may use up too many resources on some systems. One
        can control the number of workers by setting the `MAX_JOBS` environment
        variable to a non-negative number.
    r