flame on
@ 7sharp9 · Sunday, May 29, 2016 · 3 minute read · Update at May 29, 2016

A few weeks back I posted on Twitter that I was experimenting with flame graphs, In this post I will share how this was accomplished.


First of all the requirements, I’m assuming your using a Mac just as I am. If you are not then you might be able to use x-perf for windows check out summarizing-xperf-cpu-usage-with-flame-graphs for information in that area. I don’t really use Windows that often but if I do happen to try this out on Windows then I’ll pop back here and update this post.

OK, so back to requirements, a Mac, Mono Installation, Xcode installed so that you can use Instruments to collect trace information, clone the FlameGraph repo itself:

git clone https://github.com/brendangregg/FlameGraph

The FlameGraph repo is a bunch of scripts to help process the trace data and produce an svg. You can read more about FlameGraphs here: http://www.brendangregg.com/flamegraphs.html

AOT the framework

Next step is to AOT compile all the mono runtime assemblies, you can do this by running the following commands from your mono installation. For me this would be:

cd /Library/Frameworks/Mono.framework/Versions/4.4.0/lib/mono/
for i in `find gac -name '*dll'` */mscorlib.dll; do
   mono --aot $i

AOT Your app

Now you need to AOT you applications files with the same command

mono --aot myApp.exe

Attaching Instruments

Now you are ready to run your app and attach Instruments or have Instruments launch your app.

For simplicity I opted to add this to the beginning of my app:

printfn "Press any key to start"
Console.ReadKey() |> ignore

That way I could just launch my app (noting the process id) and use the process browser within Instruments to attach.
Now launch Instruments and select the Time Profiler template:

You can tweak the sampling interval with the settings on the right, in my example below I was using 40us because it was a really fast executing demo.

Use the process browser in Instruments to choose the mono process running my app e.g. mono (16314).

Now that instruments is attached hit the big red record button and hit any key on you app to start collecting data. When you are finished just hit the top button in Instruments.

You should end up with something similar to this, I used a few cycles of my 68000 emulator to get this data:

Exporting The Data

Exporting the data is pretty easy, use expand all on a node in the collected data using Cmd cLick, you can also add filters to the data, I used Atari in the screen-shot above to constrain the output to nodes that contained Atari. Now select export from the instrument menu:

Producing the FlameGraph

Now you can open a move to the flamegraph repo that you cloned earlier and execute the following command replacing myoutput.csv|svg with your input/outputs.

./stackcollapse-instruments.pl myoutput.csv | ./flamegraph.pl >myoutput.svg

You should now have a funky FlameGraph!

You could quite easily post process the csv output to clean up the mangled names that are a result of the AOT process.

Until next time …

Dave Thomas
7sharp9's blog
A cup of tea would restore my normality. - Douglas Adams
agents ast async concurrency csharp elixir elm fable fsharp functional-programming fuse games ios javascript mac mailboxprocessor metaprogramming microsoft mono monodevelop monogame monotouch myriad performance quotations running sockets threading tpl tpl-dataflow typeproviders videos xamarin youtube

© 2011 - 2020 7sharp9

Powered by Hugo with theme Dream.

About Me

Hi, my name is Dave Thomas and Im a software engineer from the Uk. I work primarily in F# but also like to tinker with pretty much any interesting language. Im drawn to more functional languages like F#, Swift and Rust.

Have questions or suggestions? Feel free to ask me on Twitter.

Thanks for reading!