ddm.partition

Functions

compute_dims(nnodes, gridsizes[, ...])	With the aim of distributing a multi-dimensional array on a Cartesian topology, compute the number of processes along each dimension.
compute_dims_general(mpi_size, npts[, ...])
compute_dims_uniform(mpi_size, npts)
partition_procs_per_patch(npts, size)	Compute the number of processes in each patch and assign to it an ascending range of processes.

Details

compute_dims(nnodes, gridsizes, min_blocksizes=None, mpi=None, try_uniform=False, mpi_dims_mask=None)

With the aim of distributing a multi-dimensional array on a Cartesian topology, compute the number of processes along each dimension.

Whenever possible, the number of processes is chosen so that the array is decomposed into identical blocks.

Parameters:

nnodesint

Number of processes in the Cartesian topology.

gridsizeslist of int

Number of array elements along each dimension.

min_blocksizeslist of int

Minimum acceptable size of a block along each dimension.

try_uniform: bool

try to decompose the array uniformly.

mpi_dims_mask: list of bool

True if the dimension is to be used in the domain decomposition (=default for each dimension). If dim_mask[i]=False, the domain decomposition yields blocksizes[i]=gridsizes[i] along the i-th dimension.

Returns:

dimslist of int

Number of processes along each dimension of the Cartesian topology.

blocksizeslist of int

Nominal block size along each dimension.

partition_procs_per_patch(npts, size)

Compute the number of processes in each patch and assign to it an ascending range of processes. The processes are distributed porportionally to the patch grid size.

Parameters:

nptslist

Number of points along each dimension for each patch.

sizeint

Number of processes.

Returns:

sizeslist of int

Number of processes in each patch.

ranges: list of list of int

The assigned ascending range of processes for each patch, the range is represented by a list of ints of size 2 [k1,k2], such that k1<=k2.