# Graupel in GT4Py

- Shaped by: Anurag
- Appetite: full cycle
- Developers: Chia Rui
- Support: Niki

## Goals
Port the sedimentation and/or transfer rate calculations while learning GT4Py.

Try to get the structure of the scheme ready in this cycle. Full implementation and verification can be done in the following cycle.

## Appetite
Entire cycle for Chia Rui.

## Solution
- [x] Chia Rui has a refactored and verified Fortran version of the scheme that aim to port. Use the implementation by David (of the original code) as the starting point.
- [ ] ~~Split the code into 3 blocks: (i) sedimentation (ii) transfer rate calculations, (iii) update the prognostics.~~
- [x] The sedimentation cannot be extracted from the graupel scheme to become a standalone piece of code because it is partially implicit. The entire scheme has to be put in one scan operator. Give adoptability priority over the performance.

- [ ] Some concerns about code structure, readability, and performance.
- [ ] Some input or output variables only has (i,j) indices (CellDim) in the Fortran code, but scan_operator along the k axis requires that all variables are either a constant or have KDim.
- [ ] Make tuning parameters constants.

FORTRAN CODE:
graupel_setup (tune_zceff_min, tune_v0snow, tune_zvz0i, tune_mu_rain, tune_rain_n0_factor, tune_icesedi_exp)
IF (PRESENT(tune_zceff_min)) THEN
M_ceff_min = tune_zceff_min
ELSE
M_ceff_min = 0.075_wp ! COSMO default
ENDIF

PYTHON CODE:
class GraupelGlobalConstants(FrozenNamespace):
GrConst_ceff_min = 0.075
GrConst_v0snow = 20.0
GrConst_vz0i = 1.25
GrConst_mu_rain = 0.0
GrConst_rain_n0_factor = 1.0
GrConst_icesedi_exp = 0.33

graupel_const : Final = GraupelGlobalConstants()

- [ ] The function Make_normalized is ignored.

#ifdef _OPENACC
! GPU code can't flush to zero double precision denormals
! So to avoid CPU-GPU differences we'll do it manually
FUNCTION make_normalized(v)
!$ACC ROUTINE SEQ
REAL(wp) :: v, make_normalized

IF (ABS(v) <= 2.225073858507201e-308_wp) THEN
make_normalized = 0.0_wp
ELSE
make_normalized = v
END IF
END FUNCTION
#else
FUNCTION make_normalized(v)
REAL(wp) :: v, make_normalized
make_normalized = v
END FUNCTION
#endif
- [ ] Super long list of type declaration in scan operator.
- [ ] Recursion depth limit. Users need to manually set the limit in the code.
- [ ] Long compilation time.

## Next task
- [ ] Prepare a test to ensure this piece of code produce the same result as the Fortran version in the next cycle.

## Rabbit holes

## No-gos
<!-- Anything specifically excluded from the concept: functionality or use cases we intentionally aren’t covering to fit the ## appetite or make the problem tractable -->