Can NV_path_rendering read data form CUDA buffers or OpenGL buffer objects?

I have a path with a huge amount of control points whose coordinates are generated by a complex transformation. The transformation process can be accelerated greatly if it runs on GPU. But glPathCommandsNV seems to be able to read from system main memory only, which requires an extra copy step from GPU memory to main memory. Considering that those coordinates data must be transferred to GPU for path rendering, there are actually two extra copy steps.

If there’s some way to specify path data with a GPU buffer, those copy steps could be eliminated, but I haven’t found any useful information on this topic.

Anyone has any ideas?