I'm considering implementing some parts of cl_intel_unified_shared_memory inside rusticl as it overlaps in a few places with cl_ext_buffer_device_address, which I already have an MR for.
Do we have any reliable tests I can develop against to see if I'm doing it correctly?
AdaptiveCPP needs it (though I prefer them supporting cl_ext_buffer_device_address instead or in addition if SVM semantics aren't needed)
