When can I trust cost_analysis

[mrlazy1708]

I've been trying to get the exact FLOP usage of a jitted function. As being documented, one way would be calling cost_analysis on a lowered function. But as also mentioned in the doc, it is not guaranteed to give consistent results. As far as I'm aware, cost of while loops (or even lax.map) can't be determined statically, and thus shouldn't be used if some code has loops. I'm wondering if there's a way to determine if the cost analysis is reliable for a given function.

Thanks!
Best