Closed pratyai closed 4 months ago
Thanks @pratyai ! Could you share a screenshot of what the final TimerOutputs displays to after this change please?
I'm trying to understand the interplay between the "Newton"
and "KKT"
tags in the @timeit
macros, and how it will interact with the timing report of solve_newton_system!
later (which is also tagged as "Newton"
in @timeit
)
Previously, at every step, one kkt.solve!()
call was under-reported (e.g. in the picture it increases by exactly 12, after I add the @timeit
call).
The "KKT" stats are nested inside the "Newton" stats, which is why I had to do the nesting with @timeit
macro, so that it doesn't get reported separately (although perhaps it should be).
Since the
KKT.solve!()
is typically one of the more expensive parts of a Newton step computation, we should account for the ~20% of calls and runtime that are currently unreported.