Optional Modules

The production validation path is ATM-focused, but the repository also contains optional modules controlled by CMake options and runtime configuration. The GPU porting policy is to GPU-enable loop-heavy timestep work, while allowing I/O and one-time setup to remain CPU-side when it does not affect timestep performance.

Module / Option	Main Files	GPU Policy
USE_HIT	hit_inflow.f90, hit_inflow_gpu.f90	GPU helper retained for runtime inflow work; setup can remain CPU
USE_SCALARS	scalars.f90	Loop-heavy scalar transport paths should remain GPU-enabled when option is active
USE_TURBINES	turbines.f90, turbines_gpu.f90, turbine_indicator.f90	Runtime turbine loops and indicator operations GPU-enabled where relevant; initialization may remain CPU
USE_LVLSET	level_set.f90, trees_*_ls.f90, level_set_base.f90	Timestep-level level-set loops require GPU coverage; tree/fmask preprocessing is setup unless repeated
USE_CPS	concurrent_precursor.f90	MPI/precursor coordination remains mostly communication/control logic
USE_CGNS	io.f90, CGNS output paths	Output dependency and I/O path; not a timestep compute target
USE_DYN_TN	Lagrangian dynamic model files	Dynamic timescale update should use existing GPU/Lagrangian paths if active

When enabling an optional module for a new case, build with the CMake option enabled, run a minimal smoke case, check the file audit for CPU-heavy uncovered loops, add GPU coverage only for repeated runtime cost, and document any intentionally CPU initialization or I/O loops.