cancel
Showing results for 
Search instead for 
Did you mean: 

Server Gurus Discussions

µProf not showing accurate callstacks when VTune does

Hi, I'm struggling to get µProf to show accurate call stacks in the call graph view. It just goes straight to [ROOT] for certain functions that I know have fairly deep (but not ridiculous) call stacks. I'm taking a simple timer-based profile on a Ryzen 7 3800X running Ubuntu 18.04. I've enabled the CSS option to capture call stacks. The app is a C++ one built with GCC 8.4.0. The project consists of multiple binaries - a main executable and other shared libraries in .so form. They are built using -O3 -g so should have debug info. I've also tried enabling -fno-omit-frame-pointers, and toggling the FPO option in µProf's capture settings, plus increasing the call stack size setting to its maximum 8 KB, to no avail.

Intel VTune captures the call stack perfectly when running exactly the same binaries, so the information is there somewhere. Is there a bug in µProf in this area? Something else I can do to narrow down the problem?

0 Likes
1 Reply
swarup
Staff

Hi @Myrmecophagavir ,

It is always preferable to use the FPO option in the uProf config with stacksize > 256B. We have made few fixes in CSS in the next release, which might be out soon. Till then, if you can share a small sample program, we would look into it and we will make sure it works as expected.

0 Likes