This isn't on this slide necessarily, but I was a little confused by the term "kernel" used in lecture. It sounds like a kernel is just a function provided by the application to the GPU, but could someone go into a little more depth? Why does it share the same name as an OS kernel, and is that intentional?

I think it just indicates that this is the "core" program for each stream processor in the GPU. Multiple programs can be running simultaneously on the GPU so each kernel is the core for each of these programs.

Just for clarification, is it correct that the memcpy primitives copy data between host memory space to device "global" memory space? If so, is it possible to directly copy host memory space data to share or private device memory?