Sparse updates on a GPU index

Hi,

I opened an issue a month ago (https://github.com/facebookresearch/faiss/issues/161) about sparse updates (i.e. updating a small number of vectors from an already trained index).

The FAISS team very nicely emitted a patch for CPU index but I would be interested to know how it could be used for GPU index.

The signature of the function I'm thinking about would look like:
`void update_vector(IVFFlat & ivf, idx_t i, const float *new_val);`

Ideally, `idx_t` i could also be a list of indices.

Is there a simple way to adapt the C++ code to modify the CUDA code?

Thanks a lot in advance for your help,

Best,
Daniel