Memory copy between two CUDA contexts

Hi all!

I have two CPU threads, each have associated CUDA context. Both work with one CUDA device.
Is it possible to copy memory from one cuda context to another cuda context ? (device to device)

I can workaround this by copying memory device to host on one context and then host to device on another context, but this is inefficient.

At the moment, sending data through an intermediary host buffer is the only way, but I think NVIDIA is working on this for a future CUDA release.