hysop.core.mpi.redistribute module

Implementation for data transfer/redistribution between topologies

`.. currentmodule : hysop.core.mpi.redistribute

See hysop.operator.redistribute.Redistribute for automatic redistribute deployment.

• RedistributeOperatorBase abstract base class

• RedistributeIntra for topologies/operators defined inside the same mpi communicator

• RedistributeInter for topologies/operators defined on two different mpi communicator

• RedistributeOverlap for topologies defined inside the same mpi parent communicator and with a different number of processes

class hysop.core.mpi.redistribute.RedistributeInter(other_task_id=None, **kwds)

Bases: RedistributeOperatorBase

Data transfer between two operators/topologies. Source and target must:

*be CartesianTopology topologies with the same global resolution *be defined on different communicators

Data transfer between two operators/topologies. Source and target must:

*be CartesianTopology topologies with the same global resolution *be defined on different communicators

apply(**kwds)

Abstract method that should be implemented. Applies this node (operator, computational graph operator…).

classmethod can_redistribute(source_topo, target_topo, other_task_id=None, **kwds)

Return true if this RedistributeOperatorBase can be applied to redistribute a variable from source_topo to target_topo, else return False.

discretize()

By default, an operator discretize all its variables. For each input continuous field that is also an output field, input topology may be different from the output topology.

After this call, one can access self.input_discrete_fields and self.output_discrete_fields, which contains input and output dicretised fields mapped by continuous fields.

self.discrete_fields will be a tuple containing all input and output discrete fields.

Discrete tensor fields are built back from discretized scalar fields and are accessible from self.input_tensor_fields, self.output_tensor_fields and self.discrete_tensor_fields like their scalar counterpart.

get_field_requirements()

Called just after handle_method(), ie self.method has been set. Field requirements are:

1. required local and global transposition state, if any.

2. required memory ordering (either C or Fortran)

Default is Backend.HOST, no min or max ghosts, MemoryOrdering.ANY and no specific default transposition state for each input and output variables.

get_preserved_input_fields()

This Inter-communicator redistribute is preserving the output fields. output fields are invalidated on other topologies only if field is not also an input

output_topology_state(output_field, input_topology_states)

Determine a specific output discrete topology state given all input discrete topology states.

Must be redefined to help correct computational graph generation. By default, just return first input state if all input states are all the same.

If input_topology_states are different, raise a RuntimeError as default behaviour. Operators altering the state of their outputs have to override this method.

The state may include transposition state, memory order and more. see hysop.topology.transposition_state.TranspositionState for the complete list.

class hysop.core.mpi.redistribute.RedistributeInterParam(parameter, source_topo, target_topo, other_task_id, domain, **kwds)

Bases: ComputationalGraphOperator

parameter transfer between two operators/topologies. Source and target must:

*be MPIParams defined on different communicators

Communicate parameter through tasks

parameter

parameter: tuple of ScalarParameter or TensorParameter

parameters to communicate

source_topo: MPIParam target_topo: MPIParam

apply(**kwds)

Abstract method that should be implemented. Applies this node (operator, computational graph operator…).

classmethod supports_mpi()

Return True if this operator was implemented to support multiple mpi processes.

class hysop.core.mpi.redistribute.RedistributeIntra(**kwds)

Bases: RedistributeOperatorBase

Data transfer between two operators/topologies. Source and target must:

*be CartesianTopology topologies with the same global resolution *be defined on the same communicator *work on the same number of mpi processes

Data transfer between two operators/topologies defined on the same communicator

Source and target must:

*be defined on the same communicator *work on the same number of mpi process *work with the same global resolution

apply(**kwds)

Abstract method that should be implemented. Applies this node (operator, computational graph operator…).

classmethod can_redistribute(source_topo, target_topo, **kwds)

Return true if this RedistributeOperatorBase can be applied to redistribute a variable from source_topo to target_topo, else return False.

discretize()

By default, an operator discretize all its variables. For each input continuous field that is also an output field, input topology may be different from the output topology.

After this call, one can access self.input_discrete_fields and self.output_discrete_fields, which contains input and output dicretised fields mapped by continuous fields.

self.discrete_fields will be a tuple containing all input and output discrete fields.

Discrete tensor fields are built back from discretized scalar fields and are accessible from self.input_tensor_fields, self.output_tensor_fields and self.discrete_tensor_fields like their scalar counterpart.