FFmpeg/dav1d's assembly isn't fully stripped, we keep local labels, which sometimes results in spam for long functions and weird perf results, since perf doesn't collect per-function stats, only per-label.
@wolf480pl If you're using perf. But if you're optimizing a function yourself as you're writing it, you'll have a much better overall function performance metric.
It's just fun to open up perf top on a system with plenty of debugging symbols to see what's happening.
@lynne so having local labels gives you more insight into which of the different parts of the function is slow, but at the same time, makes it more difficult to see the performance of the whole function?