After bug #909, we can attach to various hardware performance counters
in overflow/callback mode.  We lack a mechanism to read snapshots of
those values at random times, which would enable simple per-process or
per-cpu counting of those events.

Some syntax options, depending on bug #11550 dynamic-parameters:

global var1
probe perf.type(NN).config(MM).count(var1) {}
probe end { println("Total: ", var1) }

global var2
probe process(....).perf.type(NN).config(MM).count(var2) {}
probe end {
  foreach (v = [tid+] in var2) 
    println ("thread ", tid, " count ", v)
}