What shoc_in_and_out.cpp does
The standalone C++ driver is the equivalent of the E3SM l_turb_standalone path:
it feeds the same inputs to the same shoc_main, writes the same .txt columns, and skips the host machinery.
Line numbers on branch maint-3.0-scm-mchinita-shoc_in_and_out_cpp.
Driver shoc_in_and_out.cpp vs E3SM with l_turb_standalone
In E3SM the inputs are built from state1 first (:898) and only then overwritten by the IC files (:986); both write the same shoc_output columns; both run in exp1/exp2 modes. The driver loads the ICs directly and equals the E3SM path. The frozen-host-state integration is also what surfaced the theta-v staleness bug.
Driver ≈ the l_turb_standalone path (same inputs, same shoc_main, same .txt columns)
Real E3SM · l_turb_standalone = .true.
shoc_intr.F90 (full host interface)
1st build inputs from state1 (:898–899)
thv, thlm, qw from state1%t, state1%q
2nd standalone ONLY: OVERWRITE (:986)
read_and_set_input_to_shoc_main (sub :1448)
thv,thlm,qw,tke,u,v + state1 pressures ← ICs
3rd LOOP shoc_main × nadv (:1034, shoc.F90)
← SHARED physics, identical →
host machinery (present, but...):
ptend · energy fixer → host_dse only · outfld
standalone: does NOT feed back to
SHOC prognostics → irrelevant
Output: written by shoc_intr, after each call (:1064)
shoc_output_nadv<N>_..._nstep<S>.txt (:999)
Net effect: the 1st-step build is discarded by the
2nd-step overwrite → what reaches shoc_main = ICs.
Driver · shoc_in_and_out.cpp
replaces shoc_intr with minimal glue
load ready-made inputs from IC (:193)
thv, thlm, qw, tke read straight from file
no build step, no overwrite
the ICs already ARE the inputs — a single step
LOOP shoc_main × nadv (:317, shoc.F90)
← SAME call, same file →
host machinery:
— OMITTED —
no ptend, energy-fixer, outfld, state_copy
(unnecessary in standalone)
Output: write_state → .txt directly (:223)
<prefix>_<engine>_<mode>.txt (:304)
Same columns as E3SM (u,v,tke,qv,qc,T,P per
level & time) → the two diff directly (compare_inout.py).
≡
Two run modes — available in BOTH the driver and real E3SM
exp2 · OUTER loop: nadv = 1, shoc_main called N times → output after EACH call (driver write_state :318 · E3SM shoc_intr :1064)exp1 · INNER loop: nadv = N, shoc_main called once → E3SM writes per substep from inside shoc_main (shoc.F90 :708)exp1 ≡ exp2 bitwise: the call count changes only the output granularity, not the result.
→ Net result: driver (cxx and f90) ≡ E3SM l_turb_standalone run to ~1e-6 (compiler noise).
shoc_main (shoc.F90) is called in full and unchanged on both sides — the physics is 100% identical; only the surrounding glue differs.
A second thing this framework surfaced — the θv-staleness bug:
by integrating SHOC standalone with a frozen host state, the in-and-out framework amplified the θv-staleness
into a visible artifact — which is how the bug was identified.