Total hit count displayed by "opannotate -a" does not match sum of counts for each of the bundles in function (total count exceeds the sum by approx. 20%).
However, running the same workload under HP Caliper profiler shows almost the same total hit count, but different counts for some of the bundles (those counts are greater than the corresponding counts listed by opannotate).
Therefore, I suspect that oprofile somehow missed many hits on those bundles.
I would also like to note that such bundles have stop bit in the middle of the bundle -- is it possible that oprofile does not handle that properly?
Attaching test case assembly; source, opannotate -a output and caliper output will follow.
Log in to post a comment.