Add performance monitoring for PSyclone itself?

[arporter]

We record the performance of the code processed by PSyclone but I think it would also be useful to keep an eye on the performance of PSyclone itself - i.e. how long it takes to process/build NEMO and how long it takes to process/build LFRic.
What do @sergisiso, @LonelyCat124 and @hiker think?

[arporter]

(To be clear: I meant in the integration tests.)

[sergisiso]

I agree it would be useful.

Nemo already reports them at the end of the build step:

There is a maybe more interesting value going up the output for just the psycloning step. Are you suggesting getting those values and storing them in the database of integration test results?

[arporter]

Yes, something like that. We want to be able to keep an eye on the cost of PSyclone processing.

[LonelyCat124]

I think it might also be worth having some more detailed information on what sections are expensive for our "standard" runs, e.g. are we still accurate saying its the parser or is something like node.depth getting called so much it begins to be expensive?

[hiker]

The fab build (which will be used when adding the gungho extraction tests to the CI) will also report time for each phase, including python timing.

[hiker]

FYI, here the part of the json statistics:

        "steps": {
                "grab_folder": 0.05174525300003552,
                "find_source_files": 0.016259750000017448,
                "configurator": 3.3473102240000117,
                "c_pragma_injector": 0.010136751000004551,
                "preprocess_c": 0.010879782999950294,
                "preprocess_fortran": 6.645971768999971,
                "psyclone": 254.70674601400015,
                "analyse": 53.78163990600001,
                "compile_c": 0.0007150609999371227,
                "compile_fortran": 165.67195685499996,
                "link_exe": 9.561848357999907,
                "cleanup_prebuilds": 0.017837078000184192
        },

(note that the analyse step will basically run fparser on all Fortran files). If the required plotting lib is installed, Fab will also produce a pie chart.

[sergisiso]

Out of curiosity, have you tried the "--enable-cache" flag with the extraction. A very expensive part is the "collect_all_required_modules" extraction method that repeatedly parses the same infrastructure files each time