Effect of subroutine timing on performance

[Tom Hill (tom-j-h)]

Since LFRic Apps 3.1 compilation of subroutine timing code has been optional. We've just merged https://github.com/MetOffice/mo-bundle/pull/1096 to turn this on by default (specifically the legacy subroutine timing system) in our JEDI stack builds, but with the option to turn it off. Should we be turning it off for optimised builds (e.g., for trialling), or is the performance impact negligible?

[Andrew Coughtrie (andrewcoughtrie)]

For performance builds it is suggested that the application is built without any timing code included, if you turn it off (i.e. don't set any of the Make variables) then the LPROF logical becomes a parameter and is set to false, the optimisations in the compiler should then remove the if(LPROF) start_timing(...) lines from the code at compile time as they can never be reached. How much performance you gain from this I don't know (modern processors are very good at branch prediction) but at least you know you are never even checking the logic since the if statements aren't there.