Core performance and DrHook performance matching

[mo-mglover]

I'm keen to maintain same-or-better performance compared with DrHook; Profiler should not incur greater performance overheads. At least, not with some basic set of runtime options; it may be that more detailed analysis - such as call-tree sensitive timings - might push up the cost.

During some initial testing with a tight loop, I noticed that the traceback code in particular seemed to slow things down considerably.

There are really two parts here:

• Review/replace the C++ that tracks the traceback / call sequence.
• Incorporate performance comparisons with DrHook into the test framework.

[andrewcoughtrie]

I imagine incorporating the use of DrHook into the test framework would be difficult due to the licence, unless you mean to just include some timings for a specific set of source done offline and included for comparison?