Sends values across different processors. More...

#include <arcane/core/ITransferValuesParallelOperation.h>

Inheritance diagram for Arcane::ITransferValuesParallelOperation:

Collaboration diagram for Arcane::ITransferValuesParallelOperation:

Public Member Functions
virtual	~ITransferValuesParallelOperation ()=default
	Destructor.
virtual IParallelMng *	parallelMng ()=0
	Associated parallelism manager.
virtual void	setTransferRanks (Int32ConstArrayView ranks)=0
	Positions the array indicating who to send the values to.
virtual void	addArray (Int32ConstArrayView send_values, SharedArray< Int32 > recv_value)=0
	Adds an array of `Int32`.
virtual void	addArray (Int64ConstArrayView send_values, SharedArray< Int64 > recv_values)=0
	Adds an array of `Int64`.
virtual void	addArray (RealConstArrayView send_values, SharedArray< Real > recv_values)=0
	Adds an array of `Int64`.
virtual void	transferValues ()=0
	Sends and receives values.

Detailed Description

Sends values across different processors.

This operation allows values to be communicated with other processors. The array ranks indicates for each element the rank of the processor it is intended for. It is then possible to specify arrays containing the values to send and to receive. The send arrays must have the same number of elements as ranks

An instance is used only once. Once the transfer is complete, it can be destroyed.

For example, for a case with 3 processors:

// Processor of rank 0:
Int32UniqueArray ranks;
ranks.add(2); // Sends to rank 2
ranks.add(1); // Sends to rank 1
ranks.add(1); // Sends to rank 1
Int32UniqueArray values_1;
values_1.add(5); // Sends 5 to rank 2 (ranks[0])
values_1.add(7); // Sends 7 to rank 1 (ranks[1])
values_1.add(6); // Sends 6 to rank 1 (ranks[2])
Int64UniqueArray values_2;
values_2.add(-5); // Sends -5 to rank 2 (ranks[0])
values_2.add(-7); // Sends -7 to rank 1 (ranks[1])
values_2.add(-6); // Sends -6 to rank 1 (ranks[2])
 
// Processor of rank 1:
Int32UniqueArray ranks;
ranks.add(0); // Sends to rank 0
ranks.add(2); // Sends to rank 2
Int32UniqueArray values_1;
values_1.add(1); // Sends 1 to rank 0 (ranks[0])
values_1.add(3); // Sends 3 to rank 2 (ranks[1])
Int64UniqueArray values_2;
values_2.add(23); // Sends 23 to rank 0 (ranks[0])
values_2.add(24); // Sends 24 to rank 2 (ranks[1])
 
// Processor of rank 2:
Int32UniqueArray ranks;
ranks.add(0); // Sends to rank 0
ranks.add(0); // Sends to rank 0
Int32UniqueArray values_1;
values_1.add(0); // Sends 1 to rank 0 (ranks[0])
values_1.add(4); // Sends 3 to rank 0 (ranks[1])
Int64UniqueArray values_2;
values_2.add(-1); // Sends -1 to rank 0 (ranks[0])
values_2.add(4); // Sends 4 to rank 0 (ranks[1])

To perform the transfer

Int32UniqueArray recv_values_1;
Int64UniqueArray recv_values_2;
op->setTransferRanks(ranks);
op->addArray(values_1,recv_values_1);
op->addArray(values_2,recv_values_2);
op->transferValues();

After sending, processor of rank 0 will have the following values:

recv_values_1[0] == 1; // sent by rank 1
recv_values_1[1] == 0; // sent by rank 2
recv_values_1[2] == 4; // sent by rank 2
recv_values_2[0] == 23; // sent by rank 1
recv_values_2[1] == -1; // sent by rank 2
recv_values_2[2] == 4; // sent by rank 2

Note that the order of elements is undetermined

Definition at line 105 of file ITransferValuesParallelOperation.h.