ppmd.coulomb.octal module

class ppmd.coulomb.octal.EntryData(tree, ncomp, dtype=<class 'ctypes.c_double'>)

Bases: object

add_onto(octal_data_tree)

add data onto a OctalDataTree of mode=’halo’ :param octal_data_tree:

extract_from(octal_data_tree)

Extract data from a OctalDataTree of mode=’plain’ :param octal_data_tree:

zero()

class ppmd.coulomb.octal.OctalDataTree(tree, ncomp, mode=None, dtype=<class 'ctypes.c_double'>)

Bases: object

halo_exchange_level(level)

num_cells(start=0, end=None)

zero()

Zero all stored data.

class ppmd.coulomb.octal.OctalGridLevel(level, parent_comm, entry_map=None)

Bases: object

free()

global_to_local

Map from global cube index to local cube index

global_to_local_halo

Map from global cube index to local cube index with halos

global_to_local_parent

Map from global cube index to local parent index

grid_cube_size

Size of grid plus halos

halo_exchange(arr)

Halo exchange the passed numpy array of shape self.grid_cube_size. :param arr: numpy array to exchange.

local_grid_cube_size

Size of grid

local_grid_offset

Offset in global coodinates to the local sub-domain

owners

Map from global cube index to owning MPI rank

parent_local_size

Size of parent grid with matching ownership

class ppmd.coulomb.octal.OctalTree(num_levels, cart_comm)

Bases: object

free()

ppmd.coulomb.octal.compute_interaction_lists(local_size)

Compute the local interaction offset lists for a local domain. Child cubes are indexed lexicographically from 0. :param local_size: tuple of local cube domain dimensions (fast to slow) :return: 8x189 INT64 offsets numpy array.

ppmd.coulomb.octal.compute_interaction_offsets(cube_index)

Compute the interaction offsets for a given child cube. Child cubes are indexed with a tuple. :param cube_index: Tuple of child cube to compute offsets for. :return: numpy array type INT64 size 189x3 of offsets.

ppmd.coulomb.octal.compute_interaction_plookup()

Compute the 8x189 lookup table to map from offset index to recomputed spherical harmonics. Phi part.

ppmd.coulomb.octal.compute_interaction_radius()

Compute the coefficients such that when multiplied by the cube with they give the radius to the box centre.

ppmd.coulomb.octal.compute_interaction_tlookup()

Compute the 8x189 lookup table to map from offset index to recomputed spherical harmonics. Theta part.

ppmd.coulomb.octal.compute_local_size_offset(owners, dims, top)

class ppmd.coulomb.octal.cube_owner_map(cart_comm, cube_side_count, group_children=False)

Bases: object

Compute the global map from cube index (lexicographic) to owning MPI rank and map to MPI ranks with sub-domains of non-empty intersection with each cube.

Assumes that in each dimension ranks are assigned equal shares of the domain.

Parameters

• cart_comm – MPI cartesian communicator.

• cube_side_count – Integer subdivision level of domain.

• group_children – Consider cube subdivision to be the final mesh l of

an octal tree and group cubes such that all children of cubes in mesh l-1 are on the same MPI rank.

cart_comm

Cartesian communicator used

static compute_grid_ownership(dims, top, cube_side_count, group_children=False)

For each dimension compute the owner and contributing ranks. Does not perform out product to compute the full map. :param dims: tuple of mpi dims. :param top: tuple of top. :param cube_side_count: int, number of cubes in each dimension :param group_children: See class doc string. :return: tuple: ranks of each cube owner, contributing ranks.

static compute_map_product_contribs(cart_comm, cube_to_mpi, dim_contribs)

Compute the full map from cube index to mpi rank owner, cube index to mpi ranks of contributors. :param cart_comm: MPI cart_comm to use for logical coord to rank conversion. Owner map per dimension, elements are ints. :param cube_to_mpi: ndims x cube_side_count tuple representing cube to owner map per dimension, elements are ints. :param dim_contribs: tuple ndims x cube_side_count tuple containing lists of contributors. :return:

static compute_map_product_owners(cart_comm, dim_owners)

Compute the full map from cube index to mpi rank owner. :param cart_comm: MPI cart_comm to use for logical coord to rank conversion. :param dim_owners: ndims x cube_side_count tuple representing cube to owner map per dimension, elements are ints. lists of contributors. :return: array, length number of cubes, cube global index to mpi rank owner.

contrib_mpi

Array containing ranks contributing to a cube

contrib_starts

Cube i has contributing coefficients from ranks x_s_i to x_s_{i-1} where s is contrib_starts and x is contrib_mpi

cube_count

Total number of cubes

cube_side_count

Side count of octal tree on domain

cube_to_mpi

Array from cube global index to owning mpi rank

cube_to_send

Array if this mpi rank contributes to cube i then x_i is owning rank of cube i else x_i = -1

ppmd.coulomb.octal.cube_tuple_to_lin_xyz(ii, n)

convert xyz tuple ii to linear index in cube of side length n. :param ii: (x,y,z) tuple. :param n: side length

ppmd.coulomb.octal.cube_tuple_to_lin_zyx(ii, n)

convert xyz tuple ii to linear index in cube of side length n. :param ii: (z,y,x) tuple. :param n: side length

ppmd.coulomb.octal.send_parent_to_halo(src_level, parent_data_tree, halo_data_tree)

Copy the data from parent mode OctalDataTree to halo mode OctalDataTree.

This function enables the movement of data up the octal tree. Data is sent from src_level to src_level - 1.

Parameters

• src_level – parent level to send into the halo level

• parent_data_tree – OctalDataTree of mode parent.

• halo_data_tree – OctalDataTree of mode halo

Returns

halo_data_tree will be modified.

ppmd.coulomb.octal.send_plain_to_parent(src_level, plain_data_tree, parent_data_tree)

Copy the data from plain mode OctalDataTree to parent mode OctalDataTree.

This function enables the movement of data down the octal tree. Data is sent from src_level to src_level + 1.

Parameters

• src_level – plain level to send into the parent level

• plain_data_tree – OctalDataTree of mode plain.

• parent_data_tree – OctalDataTree of mode parent

Returns

parent_data_tree will be modified.

ppmd.coulomb.octal.shell_iterator(width)

Iterator to loop over the outer shell of a cube. :param width: Number of cells between a centred cube and the outer most cube.

width = 0 -> a single cube.

= 1 -> all cubes in a 3x3x3 block except the very middle cube.