cl_mem How is it shared across multiple devices in the context?

Accelerated Computing CUDA CUDA Programming and Performance
1

I have a basic doubt about cl_mem thing…

Say, I have a context made out of 2 CUDA GPUs. Now I create a cl_mem structure on top of it…

Later, I issue a clEnqueueRead… request to read that buffer…

Now, how much of that buffer comes from device 0 and how much comes from device 1?