try using AMD LWP to speed up dynamic instru data gathering

_From [bruen...@google.com](https://code.google.com/u/109494838902877177630/) on November 05, 2013 12:32:19_

amd64-vol2-sys.pdf 13.4

Instead of just having perf ctrs and app has to sample, LWP has a circular
buffer that collects data and when full interrupts app.
You can set buffer to any size you want.
Currently can only count cycles, instrs retired, branches retired, dcache
misses, or you can insert your own event record into the buffer.
So maybe could use this buffer for more efficient circular buffer than
fully software-implemented: have hardware check when full, etc.

_Original issue: http://code.google.com/p/dynamorio/issues/detail?id=1310_