Getting started

Let's pretend you've written a function that feels slow. We'll profile it, read the flame chart, and find the bottleneck. No prior profiling experience required.

Step 1 — Something to profile

Paste this into a Julia session (ideally a Pluto, VS Code, or Jupyter notebook):

using ProfilePerfetto

function count_primes(n)
    count = 0
    for i in 2:n
        if is_prime(i)
            count += 1
        end
    end
    return count
end

function is_prime(n)
    n < 2 && return false
    for d in 2:(n-1)         # a deliberately naive loop
        n % d == 0 && return false
    end
    return true
end

Step 2 — Profile it

@perfetto count_primes(50_000)

That's it. Behind the scenes, @perfetto:

1. Runs your expression once at a coarse sampling rate to see how long it takes.
2. Runs it again with a sharper sampling rate so the flame chart has plenty of detail without being dominated by profiler overhead.
3. Hands the resulting samples to Perfetto and renders an interactive chart inline in your notebook.

Step 3 — Reading the flame chart

You'll see a horizontal timeline. Time flows left-to-right; each row is a stack frame. The function at the top of a stack was the one actually running when the sampler fired — that's where the time is going.

Things you can do:

In our prime-counting example, you'll see is_prime dominating the chart, with most of the time spent inside its inner for loop. That's the smoking gun: replace 2:(n-1) with 2:isqrt(n) and re-run.

Step 4 — Re-run and confirm

After your fix:

@perfetto count_primes(50_000)

The is_prime bar should shrink dramatically. Profiling isn't just for finding problems — it's also how you confirm you fixed them. 🎯

Which macro to use?

• @perfetto expr / @perfetto_view expr — profiles expr and renders the chart inline in a Pluto, VS Code, or Jupyter notebook. @perfetto is an exported alias for @perfetto_view.
• @perfetto_open expr — profiles expr and opens the chart in your default web browser. Use this in the plain REPL.

Both macros accept the same trailing key = value calibration options — see Tuning the profiler.

Profiling pre-collected data

If your workload has side effects, or you already have profile data — maybe collected with Profile.@profile directly, or from a long-running session — skip the macro and drive Profile yourself:

using Profile, ProfilePerfetto

Profile.clear()
Profile.@profile my_workload()        # runs exactly once

perfetto_view()                       # or perfetto_open() in the REPL

See Using Profile.@profile manually for the full walkthrough.

Common gotchas