contains the logic of the parallel communication between participants. More...

Classes
class	BoundM2N
	An M2N between participants with a configured direction. More...
class	DistributedComFactory
class	DistributedCommunication
	Interface for all distributed solver to solver communication classes. More...
class	GatherScatterComFactory
class	GatherScatterCommunication
	Implements DistributedCommunication by using a gathering/scattering methodology. Arrays of data are always gathered and scattered at the primary. No direct communication between secondary ranks is used. For more details see m2n/DistributedCommunication.hpp. More...
class	M2N
	M2N communication class. This layer is necessary since communication between two participants can be working via several meshes, each possibly with a different decomposition. In principle, this class is only a map from meshes to DistributedCommunications. More...
class	M2NConfiguration
	Configuration for communication channels between solvers. More...
class	PointToPointComFactory
class	PointToPointCommunication
	Point-to-point communication implementation of DistributedCommunication. More...
struct	WhiteboxAccessor
	struct giving access _useOnlyPrimaryCom More...

Typedefs
using	PtrM2N = std::shared_ptr<M2N>

Functions
void	send (mesh::Mesh::VertexDistribution const &m, int rankReceiver, const com::PtrCommunication &communication)
void	receive (mesh::Mesh::VertexDistribution &m, int rankSender, const com::PtrCommunication &communication)
void	broadcastSend (mesh::Mesh::VertexDistribution const &m, const com::PtrCommunication &communication=utils::IntraComm::getCommunication())
void	broadcastReceive (mesh::Mesh::VertexDistribution &m, int rankBroadcaster, const com::PtrCommunication &communication=utils::IntraComm::getCommunication())
void	broadcast (mesh::Mesh::VertexDistribution &m)
void	print (std::map< int, std::vector< int > > const &m)
void	printCommunicationPartnerCountStats (std::map< int, std::vector< int > > const &m)
void	printLocalIndexCountStats (std::map< int, std::vector< int > > const &m)
std::map< int, std::vector< int > >	buildCommunicationMap (mesh::Mesh::VertexDistribution const &thisVertexDistribution, mesh::Mesh::VertexDistribution const &otherVertexDistribution, int thisRank=utils::IntraComm::getRank())

Detailed Description

contains the logic of the parallel communication between participants.

Typedef Documentation

◆ PtrM2N

using precice::m2n::PtrM2N = std::shared_ptr<M2N>

Definition at line 9 of file SharedPointer.hpp.

Function Documentation

◆ broadcast()

void precice::m2n::broadcast ( mesh::Mesh::VertexDistribution & m )

Definition at line 89 of file PointToPointCommunication.cpp.

Here is the call graph for this function:

◆ broadcastReceive()

void precice::m2n::broadcastReceive	(	mesh::Mesh::VertexDistribution &	m,
		int	rankBroadcaster,
		const com::PtrCommunication &	communication = utils::IntraComm::getCommunication() )

Definition at line 74 of file PointToPointCommunication.cpp.

Here is the call graph for this function:

◆ broadcastSend()

void precice::m2n::broadcastSend	(	mesh::Mesh::VertexDistribution const &	m,
		const com::PtrCommunication &	communication = utils::IntraComm::getCommunication() )

Definition at line 61 of file PointToPointCommunication.cpp.

Here is the call graph for this function:

◆ buildCommunicationMap()

std::map< int, std::vector< int > > precice::m2n::buildCommunicationMap	(	mesh::Mesh::VertexDistribution const &	thisVertexDistribution,
		mesh::Mesh::VertexDistribution const &	otherVertexDistribution,
		int	thisRank = utils::IntraComm::getRank() )

builds the communication map for a local distribution given the global distribution.

Parameters

[in]	thisVertexDistribution	the local vertex distribution
[in]	otherVertexDistribution	the total vertex distribution
[in]	thisRank	the rank to build the map for

Returns: the resulting communication map for rank thisRank

The worst case complexity of the function is: $\mathcal{O}(p 2 (2 n))$

which consists of the computation of all intersections.

n is the number of data indices for each vector in ‘otherVertexDistribution’
p number of ranks
Note that n becomes smaller, if we have more ranks.

However, in case of a proper partitioning and communication between neighbor ranks (r), we would most likely end up with a factor r<<p $\mathcal{O}(r 2 (2 n))$