separate diagnostic fill of total tend from passing of total tend to pkg/layers #720
Conversation
The values for tottend[T,S] are calculated when (1) diagnostics of tottend[T,S] are requested outside of pkg/layers AND/OR (2) diagnostics of tottend[T,S] are requested INSIDE pkg/layers. In case of (2) inside pkg/layers, prior to being diagnosed, the tottend variables are being passed around as input arguments for subroutines layers_calc.F and layers_fluxcalc.F, prior to the final diagnostic call. Here, we introduce (a) option LAYERS_DIAG_TOTTEND in LAYERS_OPTIONS.h (default to #define) to strip all array definitions, subroutine arguments, and related calculations, thus saving memory and computed time, and (b) logical "layers_diag_tottendTS" (default to .TRUE.) to allow the filling of the layers diagnotics.
|
@jm-c I made some changes, as you will see, by introducing two new things:
The reason for 1 is explained above: even if we don't want to diagnose, we're still passing around very large arrays and computing things that are all zeros. The reason for 2, instead of just using diagnostic_is_on , is that the names of the diagnostics are not fixed, with a number "1" or "2" floating around depending on order of T/S (although it might have been hardcoded 1=T, 2=S, or the other way around... but I dont want to make a mess in diagnostics_fill_state.F to determine this string, which requiring several other pkg/layers variables.. so i make it easier by defining the associated/accompanied logical. |
Due to the way tottend variables are being used, defined, calculated, and subsequently diagnosed within pkg/layers, the best approach is to wrap it around the ifdef block as now done with previously defined LAYERS_DIAG_TOTTEND compiling option without the need of accompanied complexity of another logical. Now, outside of pkg/layers, if uselayers is true, the tottend is being computed and a call is done to pass it into pkg/layers. However, inside pkg/layers, only within #ifdef LAYERS_DIAG_TOTTEND block will any calculations and diagnostics relevant to tottend will be done.
|
@jm-c after looking more carefully at the way tottend variables are used in pkg/layers and how to call this from outside, i've simplified the problem by a. doing as you suggested in diagnostics_fill_state.F: simply hook in the additional check for useLayers. Even though this variable is defined outside of pkg/layers, a subsequent call to routine layers_fill requires the additional wrap "#ifdef ALLOW_LAYERS" . Edit: the failed readthedoc is not due to any of my changes i believe. |
|
@antnguyen13 I tried to run "testreport -devel" using gfortran but got this error when running
Could you check ? |
|
@jm-c eeks, i'm appalled by my sloppiness.. now should be ok! ps- I will ask for your help at the next devel meeting on which tot[T,S]tend to diagnose, as it turns out the one I'm trying to get correct here is useless for nlfs, and so even when pkg/layers is compiled, the tot[T,S]tend being passed in are not useful at all for closing budgets. What i'm hoping to diagnose is the tot[T,S]tend that accounts for the changing hFac such that when diagnosed at THIS timestep it is equivalent to snap shots of NEXT time-step minus THIS timestep (the type we use for closing offline budgets). |
|
2 comments:
|
- Register the use of Layers-Thermodynamics code by setting new switch "layers_useThermo",
set at compile time (CPP option LAYERS_THERMODYNAMICS) since, currently, there is no
run-time params for this ;
- Add a stop if Layers-Thermodynamics code is use without pkg/diagnostics (since all
the calls are within "IF ( useDiagnostics )" ;
Note: put new switch in (new) dedicated header file "LAYERS_SWITCH.h" to facilitate
access from outside pkg/layers (no need to include LAYERS_OPTIONS.h)
using the (new) "layers_useThermo" switch (instead of "useLayers") allow to refine the condition and only call pkg/layers S/R when Layers-Thermodynamics code is used also some minor changes that minimize differences with original version
|
@antnguyen13 I pushed my little changes. If you don't like these we could revert the 2 commits. |
What changes does this PR introduce?
Enable users to calculate the total tendencies of theta and salt and pass to pkg/layers independent of whether diagnostics TOTTTEND and TOTSTEND are filled or not.
What is the current behaviour?
The calculation of total tendencies for theta/salt and passing of these to pkg/layers will be non-zeroes only if the diagnostics to TOT[T,S]TEND are filled. If not, zeros are passed to pkg/layers, variable layers_tottend is filled with zeros inside subroutine layers_fill inside layers_save.F and diagnostics LTto[1SLT,2TH,3RHO] and LSto[1SLT,2TH,3RHO] are all zeros.
What is the new behaviour
Separate the if/def block for filling the diagnostics from the calculation of tendencies, such that non-zeros tendencies can be passed to pkg/layers independent of whether the diagnostics are filled.
Does this PR introduce a breaking change?
None.
Other information:
Suggested addition to
tag-index(To avoid unnecessary merge conflicts, please don't update
tag-index. One of the admins will do that when merging your pull request.)