pm-profiler

This is an implementation of the poor man's profiler, a sample-based profiler [1].

It works by sampling at random time interval a target executable run in gdb. Samples of the call stack are collected and plotted using FlameGraphs [2].

Feel free to contribute.

Requirements

• FlameGraphs
• Stuff you normally have on a Linux distro: gdb, python, bash.

Usage

• To see the names of the functions, the target binary may need to be compiled including the debugging symbols (-g option in gcc).
• Set appropriate parameters in commands.gdb and pm_profiler.sh as described in the comments at the beginning of each file.
• To launch the profile use:

pm_profiler.sh path-to-exec gdb-commands-file output-dir

• If you want the Flame Graph only for the main function, you can run

sed '/^main/!d' samples.folded | flamegraph.pl > flame_out_main.svg

Notes:

1. Dunlavey M. Performance tuning with instruction-level cost derived from call-stack sampling. ACM SIGPLAN Notices [Internet]. Association for Computing Machinery (ACM); 2007 Aug 1;42(8):4–8. Available from: http://dx.doi.org/10.1145/1294297.1294298
2. Gregg B. The flame graph. Communications of the ACM. 2016 May 23;59(6):48-57. Available from: https://doi.org/10.1145/2909476