It would be really nice to have a way to count unhalted core cycles that only counts the cycles while executing code in a specific segment. This would allow for performance counting at user-level even with a kernel dealing with interrupts and other threads possibly preempting the userland code.
Perhaps this already exists, or another solution is available for counting cycles at user-level on a preemptive kernel?
Sidenote: It'd also be awesome to have an edx:eax -> rax and rax -> edx:eax instruction.
-Brandon Falk
