loop_over_grid.loop_over_grid_numba

loop_over_grid.loop_over_grid_numba(row_amount, col_amount, grid, dt, met_data, t_steps_per_day, toggle_dict, parallel=False, use_mpi=False, ncores='all', dask_scheduler=None, client=None)

This function wraps timestep_loop, allowing for it to be run in parallel over an arbitrarily sized grid. The grid is flattened, so for an NxN grid you don’t need a multiple of N processors in order to use all the available cores.

If model_setup.use_numba is True, then the pure Python loop_over_grid is overloaded with this function.

Parameters:

  • row_amount (Number of rows in <grid>.)

  • col_amount (Number of columns in <grid>.t_steps_per_day)

  • grid (numba.typed.List) – Nested list containing the instances of the IceShelf class for each x and y point. Vertical (z) information is stored within each class instance.

  • met_data (numpy structured array) – Grid containing the met data associated with the model grid.

  • t_steps_per_day (int) – Number of timesteps to run each day.

  • toggle_dict (dict) – Dictionary of toggle switches to be fed into MONARCHS, that determine certain things about the model (such as whether to run certain physical processes).

  • parallel, use_mpi – Dummy arguments so that we can overload the regular loop_over_grid with this Numba implementation, since these are needed there.

  • ncores – Number of cores to use. Default ‘all’, in which case it will use numba.config.NUMBA_DEFAULT_NUM_THREADS threads (i.e. all of them that Numba can detect on the system).

  • dask_scheduler, client (Dummy arguments for compatibility)

Returns:

No need to reshape flat_grid back into np.shape(grid), as each element of flat_grid is a pointer to each element of grid, i.e. operating on flat_grid changes the corresponding element of grid

Return type:

None. The function amends the instance of <grid> passed to it.