Profile1 sometimes resets its state. In those situations, if there are outstanding "Stop" calls, it might end up that we delete the root, which crahes.
This happens semi-reliable in SM60 when starting a map (presumably because of GC needs). I think it can also happen under SM52, but haven't reproduced so far.
Since we are really interested in measuring slices, as they are what takes time, and those are guaranteed to return right away, we should simply profile that.