There may be a bug that when non-zero namelist options are used for:
config_berg_initial_uvelocity_geo
config_berg_initial_vvelocity_geo
that they may not be used correctly due to a missing halo exchange. This arose during testing of small non-zero values for these options (along with config_use_berg_velocity_solver = false), where the parallelism test would fail for differences only in uBergVelocity, vBergVelocity.
The interpolation call seaice_interpolate_cell_to_vertex is done in a block loop in the subroutine init_berg_velocity, so I believe this subroutine would need to be pulled out of that block loop.
Recording this as an issue for my own tracking.