macOS Perf Approaches

macos_perf_notes.md

Remember to compile with debug.

Use DTrace - onCPU

1. Clone https://github.com/brendangregg/FlameGraph

2. Trace command sudo dtrace -c '<command>' -o out.stacks -n 'profile-997 /execname == "<command name>"/ { @[ustack(100)] = count(); }' > /dev/null

3. Create Graph FlameGraph/stackcollapse.pl out.stacks | FlameGraph/flamegraph.pl > graph1.svg

Cons, won't tell you what took the longest, just what used the most CPU time.

Instruments.app - elapsed time

https://stackoverflow.com/questions/11445619/profiling-c-on-mac-os-x/11445777#11445777

Doesn't care whether it's IO, CPU etc, will show you where your time slow downs are. Resulting output is can be selectivley collapsed/expanded to zoom in to affected parts.

In-code for Rust

Use https://docs.rs/pprof/latest/pprof/

It's programmable for more specific investigations. Only onCPU for now.

Simple example:

let guard = pprof::ProfilerGuard::new(100).unwrap();

<do some stuff>

if let Ok(report) = guard.report().build() {
  let file = File::create("flamegraph.svg").unwrap();
  report.flamegraph(file).unwrap();
};