Array of cl_mem is possible with opencl 1.1?

Is it possible to create array of cl_mem on the host and pass it to the GPU?
It is an opencl standard question, but just in case I use NVIDIA opencl 1.1.

I know this topic is discussed many times but many times I read it is not possible and many time I read it is possible.
SO, what is the truth?

Not possible. Even two dimension arrays are impossible. AFAIR (see specification) cl_mem pointers may change during the course of execution. However, in OCL 1.2 there is something called SubBuffer, also you can just make one/few large buffer/s and use offset.