​​Initial Riem_Solver_C Mini-App Port Notes
​​GT4Py miniapp at:
​​GFDL source code at:

​
​1. Sim1_Solver inlining

​​There is no way to inline a stencil within a stencil. The relevant Fortran section of code is:
​​
​​   do k=1,km
​​      do j=js1,je1
​​         do i=is1, ie1
​​            ... 
​​         enddo
​​      enddo
​​   enddo
​​!$ACC end kernels
​​      if ( a_imp > 0.5 ) then
​​           call SIM1_solver(dt, is1, ie1, js1, je1, isd, ied, jsd, jed, km, rdgas, gama, gm2, cp2, akap, pe2,  &
​​                            dm, pm2, pem, w2, dz2, pt, ws, p_fac)
​​      endif
​​!$ACC kernels present(ms,dt,km,hs,gz,pef,pe2,pem,dz2) 
​​    do k=2,km+1
​​      do j=js1,je1
​​         do i=is1, ie1
​​            ...
​​         enddo
​​      enddo
​​    enddo
​​
​​The call is based on a run-time conditional, which in the GT4Py miniapp is converted to a compile-time conditional on a_imp.
​​
​​We can inline the sim1_solver if we wrap each interval with
​​if __INLINED(A_IMP < 0.5)
​​This is the approach I’m taking at first.
​​
​​Later, it would be nice to have the following features:
  1. ​​Allow stencils calling other stencils (or we could call this a gtscript.macro​ (function-like, but allows computations)
  1. ​​Allow compile-time conditionals OUTSIDE the computations

​
​2. Filling bet

​​Bet does not need to be filled at all levels as we do in the current sim1_solver in fv3core here https://github.com/VulcanClimateModeling/fv3core/blob/master/fv3core/stencils/sim1_solver.py#L51
​​
​​It’s only used one level at a time, so this is an equivalent and higher performing version:
​​    with computation(FORWARD):
​​        with interval(2, 1):
​​            if __INLINED(A_IMP < 0.5):
​​                bet = bb[0, 0, -1]
​​    # Line 265
​​    with computation(PARALLEL), interval(2, -1):
​​        if __INLINED(A_IMP < 0.5):
​​            gam = g_rat[0, 0, -1] / bet[0, 0, 0]
​​            bet = bb - gam
​​

​
​3. Lower-dimensional temporaries (bet)

​​bet​ can truly be defined as a 2D storage. How should this be represented in GT4Py? Will:
​​
  1. ​​the toolchain be able to determine that from the code above?
  1. ​​will there be type hints on the first declaration of a temporary bet​ that hints at the storage mask?
​​

​
​4. Temporaries across computations

​​pe​ is a temporary in the miniapp stencil, but it is filled in one computation and used in another
​​

​
​5. Numpy backend has assertion error with multiple intervals + PARALLEL execution

​​
​​  File "/Users/johannd/Code/gt4py/src/gt4py/backend/numpy_backend.py", line 124, in make_stage_source
​​    assert sorted(regions, reverse=iteration_order == gt_ir.IterationOrder.BACKWARD) == regions
​​AssertionError
​​

​
​6. Need to allow empty computations due to compile-time ifs

​​