Re: [DynInst_API:] Dyninst for dynamic analysis

[svartanov@xxxxxxxxx]

Almost precisely the opposite--you'd perform static analysis to find the most efficient way to get the full data set you want out of a given function (e.g. tracing a value as it moves from register to stack to register untouched and only collecting it via instrumentation once), and then adjust your instrumentation requests accordingly. So if you want to determine how to flip all path conditions, you'd start with control and data flow analysis that tells you as precisely as possible what the path conditions are, and then find a minimal set of points that let you probe those path conditions and find how to alter them.

Thank you, now I think I understand. Optimization after instrumentation makes no sense in this case. Optimization should be performed at the instrumentation level itself and it should be hardly dependent on what instrumentation does.

Basic block static analysis seems like a pretty big deal but will greatly reduce overhead.

Best regards,

Sergey.Â