Skip to content

This document is a WORK IN PROGRESS.
This is just a quick personal cheat sheet: treat its contents with caution!



Table of contents


Check CPU limitation of your program:

$ perf stat /path/to/program/to/run # see `man perf-stat`

for a more detailed output, but the measure will be less precise (because much more info is monitored):

$ perf stat -d /path/to/program/to/run # see `man perf-stat`
$ perf stat -dd /path/to/program/to/run # see `man perf-stat`
$ perf stat -ddd /path/to/program/to/run # see `man perf-stat`

In order to stay more precise, but less frequently, one can use the --interval option.

  • Print all symbolic event types:
$ perf list
  • Find the functions where your program takes most of its time/efforts: i.e. find the "hot spots". Perf record will only work with programs compiled ahead of time (C, C++, Rust, etc)
$ perf record --call-graph=dwarf -F 100 <command-to-run> --arguments-of-the-command-to-run
  • perf report : histogramme configurable des échantillons: par binaire source, fonction, lide de code source, etc

  • perf script -F +pid > /tmp/test.perf then load it into : (then click on "Load profile from file")

If this cheat sheet has been useful to you, then please consider leaving a star here.