Generate perf stats

[markshannon]

As well as simple timings, it would be useful to have access to the stats generated by perf.
Exactly what stats are available will depend very much on the hardware and setup, but perf can generate a lot useful information, e.g.:

• Instructions retired
• Cache misses (both I and D caches)
• Stalls.

This sort of information would be very useful when trying to improve performance.
See also, #107