Timers fixed

[amartyads]

Description

Timers were broken (by me). The timer objects created in the old simulation function needed to be moved out to facilitate the breaking of the simulation loop, and these new const timer references didn't work. Rather than fix these, I've removed the objects completely, because the timer usage was inconsistent; for example, the mpi_omp_communication timer object was available previously as well, but was never used in the simulation loop. Now all the pointer and reference objects are gone, and the timers are accessed directly. Now we have several find operations with the timer map every simstep, however we already had this before for the mpi_omp_communication timer.

Resolved Issues

Timers were broken

How Has This Been Tested?

Basic gridfiller box, examples/Mamico_couette/ls1configNoCP.xml

[FG-TUM]

and these new const timer references didn't work

Can you elaborate a bit? I have no idea what the problem was and why these changes solve them 😅

[amartyads]

Ah my bad
Basically, I'm not sure why, but using these const references wasn't actually starting or stopping the timers, the outputs at the end of simulations said that the timer values were zero. For example, right now if you run something on master, it'll say "Computation took 0 seconds" at the end. The communication timers were correct, despite an omp_mpi_communication_timer reference preexisting, because this reference was never used. The other timers all say zero, if a reference was used. Now they work again.

Not sure why they were broken, because I just changed the timer approach to be more consistent instead of looking into it further, but maybe there's some issue with the timers not being defined so early in the code?

[FG-TUM]

After having a quick look at the code, it seems the problem was that the timers were initialized in e.g. Simulation::initConfigXML() but the references were already set in the constructor of Simulation. This obviously can't work. I'm not super happy with the whole way timers are handled because accessing something that isn't "registered" (aka initialized) fails silently and the way we access them now is super cumbersome, but I can live with it...