coulomb_kmc.kmc_mpi_decomp module

class coulomb_kmc.kmc_mpi_decomp.FMMMPIDecomp(fmm, max_move, boundary_condition, cuda=False)

Bases: coulomb_kmc.kmc_fmm_common.LocalOctalBase

Class that handles the MPI decomposition and converts proposed moves into the internal representation expected by the classes that handle the direct and indirect interactions.

Parameters

• fmm – FMM instance to use.

• max_move (float) – Maximum distance of any hop.

• boundary_condition – Boundary condition to use.

free_win_ind()

Free the MPI.Win for the global to local map for FMM cells.

get_local_fmm_cell(idx)

Helper function that returns the local fmm cell linear index for a particle.

Parameters

idx – Local particle index.

get_local_fmm_cell_array(ids)

For each local id in an ctypes.c_int64 numpy array returns the linear local id in the local data structure of fmm cells.

Parameters

ids – NumPy array (INT64) of local ids.

get_win_ind()

Create and return the MPI.Win for the global to local map for FMM cells.

initialise(positions, charges, fmm_cells, ids)

Initialise the data structures for the direct interactions.

Parameters

• positions – Initial positions of charges.

• charges – Initial charge values.

• fmm_cells – FMM cells of the input charges.

• ids – Unique global ids of charges.

setup_propose(moves)

Converts a tuple of proposed moves into the internal data structure for proposed moves.

Parameters

moves – Proposed moves in tuple form. e.g. ((id, proposed_positions), …).

setup_propose_with_dats(site_max_counts, current_sites, prop_positions, prop_masks, prop_energy_diffs, prop_charges=None)

Converts proposed moves passed with the propose_with_dats interface into the internal data structure for proposed moves that can be passed to the classes for direct and indirect interactions.

Parameters

• site_max_counts – ScalarArray, dtype=c_int64 Input

• current_sites – ParticleDat, dtype=c_int64 Input

• prop_positions – ParticleDat, dtype=c_double Input

• prop_masks – ParticleDat, dtype=c_int64 Input

• prop_energy_diffs – ParticleDat, dtype=c_double Output

• prop_charges – ParticleDat, dtype=c_double Input (optional)

Returns:

tuple – (total_moves, num_particles, host_data, cuda_data), where total_moves is the total number of proposed moves, num_particles is the number of particles (charges) with proposed moves, host_data is the proposed move data on the host and cuda_data is the proposed move data on a CUDA device (if applicable).

The host_data and cuda_data contain the following entries:

rate_location	Storage location of change in energy value.
new_positions	Proposed positions.
new_charges	Proposed charges.
new_ids	Global ids of proposed moves.
new_fmm_cells	New FMM cell containing proposed move.
new_shifted_positions	Proposed position shifted into halo region on sub-domain.
old_positions	Original position.
old_charges	Original charge.
old_ids	Original global id.
old_fmm_cells	Original FMM cell.
exclusive_sum	Used as a map from old positions to new positions.