[DynInst_API:] DynInst Overhead

[Buddhika Chamith Kahawitage Don <budkahaw@xxxxxxxxxxxx>]

Hi All,

I am trying to see how much an overhead DynInst would typically impose. To test it myself I ran Dyninst instrumented h264 with spec 2006 reference inputs [1] (compiled with gcc). I was using the CodeCoverage tool [2] for instrumentation. But it's taking way too higher time than the unprofiled run. (More than a 50x slow down) although the DynInst paper [3] suggest an overhead less than 1.5x for basic block count. Not sure if I am missing something here. Any ideas on this?

Cheers
Bud

[1] https://www.spec.org/cpu2006/Docs/464.h264ref.html
[2] http://www.paradyn.org/html/tools/codecoverage.html
[3] ftp://ftp.cs.wisc.edu/paradyn/papers/Bernat11AWAT.pdf