topic Statistical call graph visualization in Analyzers

Statistical call graph visualization

dashesy — Thu, 17 May 2012 16:22:12 GMT

I am evaluatign the product for a Linux application, I am not exactly trying to improve performance, but to study the current application we have.
Reading the forum I know the call graphs are no longer exact which is fine because I have plenty of other tools for that. I think the wall-clock time is a better tool for identifying the hot spots any ways.
However, that would be nice if I could get a nice traditional diagram of the statistical call graph, then I could print it on a large paper and study the paths taken.
BTW, I can use gprof2dot application with callgrind, gprof, perf, oprofile, ... with really nice results, if there is any documentation on the saved format that could help in creating a plugin for VTune as well.
One more thing, the comparision assumes both profiles run exactly for the same duration, is there anyway to normalize the results by total time consumed?

Statistical call graph visualization

Peter_W_Intel — Fri, 18 May 2012 03:05:50 GMT

1. The user can export VTune resultin gprof format:amplxe-cl -report gprof-cc -r r000?hs
2. The user can use amplxe-gui to open result and view in Top-down Treepanel, which has a tree view with all function nodes (with total CPU time and self CPU time).

Regards, Peter

It's possible to use

Kirill_R_Intel — Fri, 05 Apr 2013 07:15:29 GMT

It's possible to use gprof2dot with VTune results - read this blog post.