NVIDIA's run-time CUDA compilation library, NVRTC, supports the compilation flag --no-source-include. The documentation describes this flag as follows:
The preprocessor by default adds the directory of each input sources to the include path. This option disables this feature and only considers the path specified explicitly.
I don't quite understand how this is meaningful for NVRTC, as opposed to a regular compiler: We feed NVRTC a single input source, as a null-terminated string; we do not provide it with the path of a source file to read. So what exactly does this flag do?