TML_Comm Class Reference

abstract base class for communicator More...

#include <comm.h>

Inheritance diagram for TML_Comm:

Public Member Functions
bool	isNull () const
int	rank () const
int	size ()
MPI_Comm	comm () const
TML_Comm &	operator= (const TML_Comm &)
void	setComm (MPI_Comm)
	TML_Comm ()
	TML_Comm (MPI_Comm)
TML_Comm	include (const vector< int > &)
TML_Comm	exclude (const vector< int > &)
template<typename T >
void	send (T, int, int=0)
template<typename T >
void	receive (T &, int, int=MPI_ANY_TAG)
template<typename T >
void	send_array (T *, int, int, int=0)
template<typename T >
void	receive_array (T *, int, int, int=MPI_ANY_TAG)
template<typename T >
void	send_cont (const T &, int, int=0)
template<typename T >
void	receive_cont (T &, int, int=MPI_ANY_TAG)
template<typename T >
void	send_cont_packed (T, int, bool, int=0)
template<typename T >
void	receive_cont_packed (T &, int, bool, int=MPI_ANY_TAG)
template<typename T , typename P >
void	sendrecv (T, P &, int, int, int=0)
template<typename T , typename P >
void	sendrecv_array (T *, int, P *, int, int, int, int=0)
template<typename T , typename P >
void	sendrecv_cont (T, P &, int, int, int=0)
template<typename T >
void	sendrecv_cont_replace (T &, int, int, int=0)
template<typename T , typename P >
void	sendrecv_cont_packed (T, P &, int, int, bool, int=0)
template<typename T >
void	sendrecv_cont_packed_replace (T &, int, int, bool, int=0)
template<typename T >
void	broadcast (T)
template<typename T >
void	broadcast_array (T *, int)
template<typename T >
void	broadcast_cont (const T &)
template<typename T >
void	broadcast_cont_packed (const T &)
template<typename T >
void	recv_broadcast (T &, int)
template<typename T >
void	recv_broadcast_array (T *, int, int)
template<typename T >
void	recv_broadcast_cont (T &, int)
template<typename T >
void	recv_broadcast_cont_packed (T &, int)
template<typename T >
void	scatter (const multimap< int, T >)
template<typename T >
void	recv_scatter (T &, int)
template<typename T >
void	gather (multimap< int, T > &)
template<typename T >
void	send_gather (T &, int)
template<typename T >
void	gather_debug (multimap< int, T > &)
template<typename T >
void	send_gather_debug (T &, int)
template<typename T >
void	scatter_packed (const multimap< int, T >)
template<typename T >
void	recv_scatter_packed (T &, int)
template<typename T >
void	gather_packed (multimap< int, T > &)
template<typename T >
void	send_gather_packed (const T &, int)
template<typename T >
T	sum_all (const T &)
void	barrier ()
void	barrier (const string &)

Protected Attributes
MPI_Status	m_status
MPI_Comm	m_comm

Detailed Description

abstract base class for communicator

Author

Steffen Abe $Revision$ $Date$

Constructor & Destructor Documentation

TML_Comm() [1/2]

TML_Comm::TML_Comm

(

)

Default constructor for TML_comm. Sets the MPI communicator to MPI_COMM_NULL;

References m_comm.

TML_Comm() [2/2]

TML_Comm::TML_Comm

(

MPI_Comm

comm

)

construct TML_Comm from MPI communicator

Parameters

comm	the MPI communicator

References comm(), and m_comm.

Here is the call graph for this function:

Member Function Documentation

barrier() [1/2]

void TML_Comm::barrier

(

)

Wait on a barrier. Wrapper for MPI_Barrier.

References m_comm.

Referenced by CLatticeMaster::addVectorWallField(), CLatticeMaster::do2dCalculations(), CSubLatticeControler::initMPI(), CLatticeMaster::makeLattice(), CSubLatticeControler::run(), CLatticeMaster::saveTimingDataToFile(), CLatticeMaster::setSpatialDomain(), CLatticeMaster::setupWorkers(), and VectorWallFieldMaster::VectorWallFieldMaster().

Here is the caller graph for this function:

barrier() [2/2]

void TML_Comm::barrier

(

const string &

msg

)

Wait on a barrier with debug message

References console, Console::Debug(), m_comm, and rank().

Here is the call graph for this function:

broadcast()

template<typename T >

void TML_Comm::broadcast

(

T

data

)

broadcast single data to all other nodes

Parameters

data	the data to be broadcast

References GetType, m_comm, and rank().

Referenced by broadcast_cont(), broadcast_cont_packed(), ScalarInteractionFieldMaster::collect(), ScalarParticleFieldMaster::collect(), VectorParticleFieldMaster::collect(), ScalarParticleDistributionMaster::collect(), ScalarTriangleFieldMaster::collect(), VectorInteractionFieldMaster::collect(), VectorTriangleFieldMaster::collect(), VectorWallFieldMaster::collect(), ScalarInteractionFieldMaster::collectFull(), ScalarParticleFieldMaster::collectFull(), ScalarTriangleFieldMaster::collectFull(), VectorInteractionFieldMaster::collectFull(), VectorTriangleFieldMaster::collectFull(), ScalarInteractionFieldMaster::collectFull2(), VectorInteractionFieldMaster::collectFull2(), ScalarInteractionFieldMaster::collectFullWithID(), VectorInteractionFieldMaster::collectFullWithID(), ScalarInteractionFieldMaster::collectFullWithPosID(), VectorInteractionFieldMaster::collectFullWithPosID(), ScalarInteractionFieldMaster::collectSum(), ScalarParticleFieldMaster::collectSum(), VectorInteractionFieldMaster::collectSum(), ScalarInteractionFieldMaster::ScalarInteractionFieldMaster(), ScalarParticleFieldMaster::ScalarParticleFieldMaster(), ScalarTriangleFieldMaster::ScalarTriangleFieldMaster(), test_broadcast(), VectorInteractionFieldMaster::VectorInteractionFieldMaster(), VectorParticleFieldMaster::VectorParticleFieldMaster(), VectorTriangleFieldMaster::VectorTriangleFieldMaster(), and VectorWallFieldMaster::VectorWallFieldMaster().

Here is the call graph for this function:

Here is the caller graph for this function:

broadcast_array()

template<typename T >

void TML_Comm::broadcast_array	(	T *	data,
		int	ndata
	)

broadcast an array of known size

Parameters

data	the array to be broadcast
ndata	the size of the array (nr. of elements)

References GetType, m_comm, and rank().

Referenced by broadcast_cont(), and broadcast_cont_packed().

Here is the call graph for this function:

Here is the caller graph for this function:

broadcast_cont()

template<typename T >

void TML_Comm::broadcast_cont

(

const T &

data

)

broadcast the content of a STL container of simple types. Uses 2 MPI broadcasts for size and data.

Parameters

data	the data to be broadcast

References broadcast(), broadcast_array(), and esys::lsm::bpu::iter().

Referenced by ScalarInteractionFieldMaster::ScalarInteractionFieldMaster(), ScalarParticleFieldMaster::ScalarParticleFieldMaster(), ScalarTriangleFieldMaster::ScalarTriangleFieldMaster(), CLatticeMaster::setSpatialDomain(), test_broadcast_cont(), VectorInteractionFieldMaster::VectorInteractionFieldMaster(), VectorParticleFieldMaster::VectorParticleFieldMaster(), VectorTriangleFieldMaster::VectorTriangleFieldMaster(), and VectorWallFieldMaster::VectorWallFieldMaster().

Here is the call graph for this function:

Here is the caller graph for this function:

broadcast_cont_packed()

template<typename T >

void TML_Comm::broadcast_cont_packed

(

const T &

data

)

broadcast the content of a STL container of packable objects. Uses 2 MPI broadcasts for size and data.

Parameters

data	the data to be broadcast

References broadcast(), broadcast_array(), TML_Packed_Message::buffer(), esys::lsm::bpu::iter(), m_comm, TML_PackedMessageInterface::pack(), and TML_Packed_Message::size().

Referenced by CLatticeMaster::createTriMesh(), CLatticeMaster::readAndDistributeMesh2D(), CLatticeMaster::setSpatialDomain(), and test_broadcast_cont_packed().

Here is the call graph for this function:

Here is the caller graph for this function:

comm()

MPI_Comm TML_Comm::comm ( ) const

inline

References m_comm.

Referenced by setComm(), test_pack(), test_pack_multi(), TML_CartComm::TML_CartComm(), and TML_Comm().

Here is the caller graph for this function:

exclude()

TML_Comm TML_Comm::exclude

(

const vector< int > &

ids

)

Construct a new communicator containing the processes from the current communicator except to ones which are given as input

Parameters

ids	the ranks of the processes which are excluded from the new communicator

Todo:

error handling

References m_comm.

gather()

template<typename T >

void TML_Comm::gather

(

multimap< int, T > &

data

)

Gather data from all other processes in the communicator into a multimap. The multimap-key will be set according to the rank of the process where the data came from

Parameters

data	the multimap

References GetType, m_comm, rank(), and size().

Referenced by VectorParticleFieldMaster::collect(), VectorWallFieldMaster::collect(), ScalarInteractionFieldMaster::collectFull(), ScalarParticleFieldMaster::collectFull(), ScalarTriangleFieldMaster::collectFull(), VectorInteractionFieldMaster::collectFull(), VectorTriangleFieldMaster::collectFull(), ScalarInteractionFieldMaster::collectFull2(), VectorInteractionFieldMaster::collectFull2(), ScalarInteractionFieldMaster::collectFullWithID(), VectorInteractionFieldMaster::collectFullWithID(), ScalarInteractionFieldMaster::collectFullWithPosID(), VectorInteractionFieldMaster::collectFullWithPosID(), ScalarInteractionFieldMaster::collectSum(), ScalarParticleFieldMaster::collectSum(), VectorInteractionFieldMaster::collectSum(), CLatticeMaster::findParticleNearestTo(), CLatticeMaster::getNumParticles(), CLatticeMaster::getParticlePosn(), CLatticeMaster::getSlaveSpatialDomains(), CLatticeMaster::getSphereBodyForce(), CLatticeMaster::getSphereBodyPosn(), CLatticeMaster::getWallForce(), CLatticeMaster::getWallPosn(), CheckPointController::issueCheckPointCmdWTM(), and test_gather().

Here is the call graph for this function:

Here is the caller graph for this function:

gather_debug()

template<typename T >

void TML_Comm::gather_debug

(

multimap< int, T > &

data

)

Gather data from all other processes in the communicator into a multimap. The multimap-key will be set according to the rank of the process where the data came from. Debug version (output data)

Parameters

data	the multimap

References console, Console::Debug(), GetType, m_comm, rank(), and size().

Here is the call graph for this function:

gather_packed()

template<typename T >

void TML_Comm::gather_packed

(

multimap< int, T > &

data

)

References TML_Packed_Message::buffer(), console, Console::Debug(), GetType, m_comm, TML_Packed_Message::pop_int(), rank(), size(), and TML_PackedMessageInterface::unpack().

Here is the call graph for this function:

include()

TML_Comm TML_Comm::include

(

const vector< int > &

ids

)

Construct a new communicator containing the processes which are given as input

Parameters

ids	the ranks of the processes which form the new communicator in the current communicator

Todo:

error handling

References console, Console::Error(), m_comm, and rank().

Referenced by test_const().

Here is the call graph for this function:

Here is the caller graph for this function:

isNull()

bool TML_Comm::isNull ( ) const

inline

References m_comm.

operator=()

TML_Comm & TML_Comm::operator=

(

const TML_Comm &

rhs

)

References m_comm.

rank()

int TML_Comm::rank

(

)

const

References m_comm.

Referenced by barrier(), broadcast(), broadcast_array(), gather(), gather_debug(), gather_packed(), TML_CartComm::get_coords(), include(), CSubLatticeControler::initMPI(), main(), scatter(), scatter_packed(), and test_const().

Here is the caller graph for this function:

receive()

template<typename T >

void TML_Comm::receive	(	T &	data,
		int	source,
		int	tag = MPI_ANY_TAG
	)

receive single data

Parameters

data	the data to be received
source	the rank of the destination process the data comes from
tag	the message tag

Warning

no checks

References GetType, m_comm, and m_status.

Referenced by receive_cont(), receive_cont_packed(), test_pair(), test_simple(), and VectorWallFieldMaster::VectorWallFieldMaster().

Here is the caller graph for this function:

receive_array()

template<typename T >

void TML_Comm::receive_array	(	T *	data,
		int	ndata,
		int	source,
		int	tag = MPI_ANY_TAG
	)

receive a C-array of data with known dimensions

Parameters

data	the data to be received
ndata	the number of integers to be received
source	the rank of the destination process the data comes from
tag	the message tag

Warning

no checks

References GetType, m_comm, and m_status.

Referenced by receive_cont(), and receive_cont_packed().

Here is the caller graph for this function:

receive_cont()

template<typename T >

void TML_Comm::receive_cont	(	T &	data,
		int	source,
		int	tag = MPI_ANY_TAG
	)

Receive an STL container or anything that has iterators, begin() and end(). Single item insert (a.insert(p,t)) is used instead of range insert to be compatible with both sequence and associative containers;

Parameters

data	the data to be received
source	the rank of the destination process the data is coming from
tag	the message tag

References receive(), and receive_array().

Referenced by test_container(), test_container_map(), and test_container_vec3().

Here is the call graph for this function:

Here is the caller graph for this function:

receive_cont_packed()

template<typename T >

void TML_Comm::receive_cont_packed	(	T &	data,
		int	source,
		bool	checked,
		int	tag = MPI_ANY_TAG
	)

Receive an STL container or anything that has iterators and insert(). Uses 2 receive operations for size and data. If the "checked" option is set, a checked message buffer is used to detect type mismatches between pack and unpack operations.

Parameters

data	the data to be received
source	the rank of the destination process the data is coming from
checked	use checked message buffer if true
tag	the message tag

Warning

checked stuff not yet implemented

References TML_Packed_Message::buffer(), m_comm, TML_Packed_Message::pop_int(), receive(), receive_array(), and TML_PackedMessageInterface::unpack().

Referenced by test_container_packed(), and test_container_packed_user().

Here is the call graph for this function:

Here is the caller graph for this function:

recv_broadcast()

template<typename T >

void TML_Comm::recv_broadcast	(	T &	data,
		int	root
	)

Receive broadcast of single data from a given node

Parameters

data	the data to be broadcast
root	the node which sent the broadcast

References GetType, and m_comm.

Referenced by recv_broadcast_cont(), recv_broadcast_cont_packed(), ScalarTriangleFieldSlave::sendData(), VectorEdge2DFieldSlave::sendData(), VectorTriangleFieldSlave::sendData(), and test_broadcast().

Here is the caller graph for this function:

recv_broadcast_array()

template<typename T >

void TML_Comm::recv_broadcast_array	(	T *	data,
		int	ndata,
		int	root
	)

Receive broadcast of an array of known size

Parameters

data	the array to be broadcast
ndata	the size of the array (nr. of elements)
root	the node which sent the broadcast

References GetType, and m_comm.

Referenced by recv_broadcast_cont(), and recv_broadcast_cont_packed().

Here is the caller graph for this function:

recv_broadcast_cont()

template<typename T >

void TML_Comm::recv_broadcast_cont	(	T &	data,
		int	root
	)

Receive broadcast of the content of a STL container of simple types from a given node. Uses 2 MPI broadcasts for size and data.

Parameters

data	the data to be broadcast
root	the node which sent the broadcast

References recv_broadcast(), and recv_broadcast_array().

Referenced by CSubLatticeControler::getIdParticleData(), CSubLatticeControler::initLatticeCirc(), and test_broadcast_cont().

Here is the call graph for this function:

Here is the caller graph for this function:

recv_broadcast_cont_packed()

template<typename T >

void TML_Comm::recv_broadcast_cont_packed	(	T &	data,
		int	root
	)

Receive broadcast of the content of a STL container of packable objects from a given node. Uses 2 MPI broadcasts for size and data.

Parameters

data	the data to be broadcast
root	the node which sent the broadcast

References TML_Packed_Message::buffer(), m_comm, TML_Packed_Message::pop_int(), recv_broadcast(), recv_broadcast_array(), and TML_PackedMessageInterface::unpack().

Referenced by CSubLatticeControler::initLattice(), CSubLatticeControler::initLatticeCirc(), and test_broadcast_cont_packed().

Here is the call graph for this function:

Here is the caller graph for this function:

recv_scatter()

template<typename T >

void TML_Comm::recv_scatter	(	T &	data,
		int	root
	)

receive scattered data

Parameters

data	the received data
root	the process which scattered the data

References GetType, m_comm, NULL, and size().

Referenced by test_scatter().

Here is the call graph for this function:

Here is the caller graph for this function:

recv_scatter_packed()

template<typename T >

void TML_Comm::recv_scatter_packed	(	T &	data,
		int	root
	)

receive scattered packed data

Parameters

data	the received data
root	the process which scattered the data

References TML_Packed_Message::begin_unpack(), TML_Packed_Message::buffer(), GetType, m_comm, NULL, TML_Packed_Message::pop_int(), and TML_PackedMessageInterface::unpack().

Referenced by test_scatter_packed().

Here is the call graph for this function:

Here is the caller graph for this function:

scatter()

template<typename T >

void TML_Comm::scatter

(

const multimap< int, T >

data

)

scatter the content of a multimap. The key of the multimap is used to decide where to send the data. Uses one MPI_Scatter (sizes) and one MPI_Scatterv (data) call.

Parameters

data	the multimap containing the data to be scattered

Todo:

checks, handling of out of range indices

References GetType, esys::lsm::bpu::iter(), m_comm, rank(), and size().

Referenced by test_scatter().

Here is the call graph for this function:

Here is the caller graph for this function:

scatter_packed()

template<typename T >

void TML_Comm::scatter_packed

(

const multimap< int, T >

data

)

Scatter the content of a multimap of packable objects. The key of the multimap is used to decide where to send the data. Uses one MPI_Scatter (sizes) and one MPI_Scatterv (data) call.

Parameters

data	the multimap containing the data to be scattered

Todo:

checks, handling of out of range indices

References TML_PackedMultiMessage::buffer(), GetType, esys::lsm::bpu::iter(), m_comm, TML_PackedMultiMessage::offsets(), rank(), size(), and TML_PackedMultiMessage::sizes().

Referenced by test_scatter_packed().

Here is the call graph for this function:

Here is the caller graph for this function:

send()

template<typename T >

void TML_Comm::send	(	T	data,
		int	dest,
		int	tag = 0
	)

send single data

Parameters

data	the data to be sent
dest	the rank of the destination process the data is sent to
tag	the message tag

References GetType, and m_comm.

Referenced by send_cont(), send_cont_packed(), test_pair(), and test_simple().

Here is the caller graph for this function:

send_array()

template<typename T >

void TML_Comm::send_array	(	T *	data,
		int	ndata,
		int	dest,
		int	tag = 0
	)

send a C-array of data with known dimensions

Parameters

data	the data to be sent
ndata	the size of the array (nr. of elements)
dest	the rank of the destination process the data is sent to
tag	the message tag

Warning

no checks

References GetType, and m_comm.

Referenced by send_cont(), and send_cont_packed().

Here is the caller graph for this function:

send_cont()

template<typename T >

void TML_Comm::send_cont	(	const T &	data,
		int	dest,
		int	tag = 0
	)

Send an STL container or anything that has iterators, begin() and end(). Uses 2 send operations for size and data.

Parameters

data	the data to be sent
dest	the rank of the destination process the data is sent to
tag	the message tag

References esys::lsm::bpu::iter(), send(), and send_array().

Referenced by test_container(), test_container_map(), and test_container_vec3().

Here is the call graph for this function:

Here is the caller graph for this function:

send_cont_packed()

template<typename T >

void TML_Comm::send_cont_packed	(	T	data,
		int	dest,
		bool	checked,
		int	tag = 0
	)

Send an STL container or anything that has iterators, begin() and end(). Uses 2 send operations for size and data. If the "checked" option is set, a checked message buffer is used to detect type mismatches between pack and unpack operations. The TML_Packed_Message has to grow dynamically. It is not possible to predetermine the size of the message because while T always has a size() function, it is not guaranteed that sizeof() or size() work for the valuetype of T.

Parameters

data	the data to be sent
dest	the rank of the destination process the data is sent to
checked	use checked message buffer if true
tag	the message tag

Warning

checked stuff not yet implemented

References TML_Packed_Message::buffer(), esys::lsm::bpu::iter(), m_comm, TML_PackedMessageInterface::pack(), send(), send_array(), and TML_Packed_Message::size().

Referenced by test_container_packed(), and test_container_packed_user().

Here is the call graph for this function:

Here is the caller graph for this function:

send_gather()

template<typename T >

void TML_Comm::send_gather	(	T &	data,
		int	root
	)

Send data to be gathered by a root process.

Parameters

data	the data (a container)
root	the root process

References GetType, esys::lsm::bpu::iter(), m_comm, NULL, and size().

Referenced by CSubLatticeControler::findParticleNearestToPoint(), CSubLatticeControler::getNumParticles(), CSubLatticeControler::getParticlePosn(), CSubLatticeControler::initLattice(), CSubLatticeControler::initLatticeCirc(), CheckPointer::saveThroughMaster(), ScalarTriangleFieldSlave::SendDataFull(), VectorEdge2DFieldSlave::SendDataFull(), VectorTriangleFieldSlave::SendDataFull(), and test_gather().

Here is the call graph for this function:

Here is the caller graph for this function:

send_gather_debug()

template<typename T >

void TML_Comm::send_gather_debug	(	T &	data,
		int	root
	)

Send data to be gathered by a root process. Debug version.

Parameters

data	the data (a container)
root	the root process

References console, Console::Debug(), GetType, esys::lsm::bpu::iter(), m_comm, NULL, and size().

Here is the call graph for this function:

send_gather_packed()

template<typename T >

void TML_Comm::send_gather_packed	(	const T &	data,
		int	root
	)

References TML_Packed_Message::buffer(), console, Console::Debug(), GetType, esys::lsm::bpu::iter(), m_comm, NULL, TML_PackedMessageInterface::pack(), size(), and TML_Packed_Message::size().

Here is the call graph for this function:

sendrecv()

template<typename T , typename P >

void TML_Comm::sendrecv	(	T	send_data,
		P &	recv_data,
		int	dest,
		int	source,
		int	tag = 0
	)

send and receive single data

Parameters

send_data	the data to be sent
recv_data	the data to be received
dest	the rank of the destination process the data is sent to
source	the rank of the destination process the data comes from
tag	the message tag

Warning

no checks

References GetType, m_comm, and m_status.

Referenced by sendrecv_cont(), sendrecv_cont_packed(), sendrecv_cont_packed_replace(), sendrecv_cont_replace(), TML_CartComm::shift(), TML_CartComm::shift_array(), and test_simple_sendrecv().

Here is the caller graph for this function:

sendrecv_array()

template<typename T , typename P >

void TML_Comm::sendrecv_array	(	T *	send_data,
		int	send_count,
		P *	recv_data,
		int	recv_count,
		int	dest,
		int	source,
		int	tag = 0
	)

send and receive a C-array of data with known dimensions

Parameters

send_data	the data to be sent
send_count	the number of integers to be sent
recv_data	the data to be received
recv_count	the number of integers to be received
dest	the rank of the destination process the data is sent to
source	the rank of the destination process the data comes from
tag	the message tag

Warning

no checks

References GetType, m_comm, and m_status.

Referenced by sendrecv_cont(), sendrecv_cont_packed(), sendrecv_cont_packed_replace(), and sendrecv_cont_replace().

Here is the caller graph for this function:

sendrecv_cont()

template<typename T , typename P >

void TML_Comm::sendrecv_cont	(	T	send_data,
		P &	recv_data,
		int	dest,
		int	source,
		int	tag = 0
	)

send and receive an STL container or anything that has iterators, begin() and end(). Uses 2 MPI_Sendrecv operations for size and data

Parameters

send_data	the data to be sent
recv_data	the data to be received
dest	the rank of the destination process the data is sent to
source	the rank of the destination process the data comes from
tag	the message tag

References esys::lsm::bpu::iter(), sendrecv(), and sendrecv_array().

Referenced by TML_CartComm::shift_cont(), and test_container_sendrecv().

Here is the call graph for this function:

Here is the caller graph for this function:

sendrecv_cont_packed()

template<typename T , typename P >

void TML_Comm::sendrecv_cont_packed	(	T	send_data,
		P &	recv_data,
		int	dest,
		int	source,
		bool	checked,
		int	tag = 0
	)

send and receive an STL container or anything that has iterators, begin() and end() and insert(). Uses 2 MPI_Sendrecv operations for size and data. If the "checked" option is set, a checked message buffer is used to detect type mismatches between pack and unpack operations.

Parameters

send_data	the data to be sent
recv_data	the data to be received
dest	the rank of the destination process the data is sent to
source	the rank of the destination process the data comes from
tag	the message tag

Warning

checked stuff not yet implemented

References TML_Packed_Message::buffer(), esys::lsm::bpu::iter(), m_comm, TML_PackedMessageInterface::pack(), TML_Packed_Message::pop_int(), sendrecv(), sendrecv_array(), TML_Packed_Message::size(), and TML_PackedMessageInterface::unpack().

Referenced by TML_CartComm::shift_cont_packed(), and test_container_sendrecv_packed().

Here is the call graph for this function:

Here is the caller graph for this function:

sendrecv_cont_packed_replace()

template<typename T >

void TML_Comm::sendrecv_cont_packed_replace	(	T &	data,
		int	dest,
		int	source,
		bool	checked,
		int	tag = 0
	)

send and receive an STL container or anything that has iterators, begin() and end() and insert(). In input data is replaced by the received data. Uses 2 MPI_Sendrecv operations for size and data. If the "checked" option is set, a checked message buffer is used to detect type mismatches between pack and unpack operations.

Parameters

data	the data to be sent and received
dest	the rank of the destination process the data is sent to
source	the rank of the destination process the data comes from
tag	the message tag

Warning

checked stuff not yet implemented

References TML_Packed_Message::buffer(), esys::lsm::bpu::iter(), m_comm, TML_PackedMessageInterface::pack(), TML_Packed_Message::pop_int(), sendrecv(), sendrecv_array(), TML_Packed_Message::size(), and TML_PackedMessageInterface::unpack().

Referenced by test_container_sendrecv_packed_replace().

Here is the call graph for this function:

Here is the caller graph for this function:

sendrecv_cont_replace()

template<typename T >

void TML_Comm::sendrecv_cont_replace	(	T &	data,
		int	dest,
		int	source,
		int	tag = 0
	)

send and receive an STL container or anything that has iterators, begin() and end(), erase() and insert(). The input data is replaced (via erase/insert) by the output data. Uses separate send/receive buffers and MPI_Sendrecv instead of a single buffer and MPI_Sendrecv_replace to accomodate different size send and received messages. Uses 2 MPI_Sendrecv operations for size and data.

Parameters

data	the data to be sent and received
dest	the rank of the destination process the data is sent to
source	the rank of the destination process the data comes from
tag	the message tag

References esys::lsm::bpu::iter(), sendrecv(), and sendrecv_array().

Referenced by test_container_sendrecv_replace().

Here is the call graph for this function:

Here is the caller graph for this function:

setComm()

void TML_Comm::setComm

(

MPI_Comm

comm

)

set the undelying MPI communicator

Parameters

comm	the MPI communicator

References comm(), and m_comm.

Referenced by CSubLatticeControler::initMPI(), CLatticeMaster::setupWorkers(), and TSubLattice< T >::TSubLattice().

Here is the call graph for this function:

Here is the caller graph for this function:

size()

int TML_Comm::size

(

)

References m_comm.

Referenced by cart_comm_build(), VectorWallFieldMaster::collect(), gather(), gather_debug(), gather_packed(), CSubLatticeControler::initMPI(), main(), recv_scatter(), scatter(), scatter_packed(), send_gather(), send_gather_debug(), send_gather_packed(), test_sum_all(), and TML_CartComm::TML_CartComm().

Here is the caller graph for this function:

sum_all()

template<typename T >

T TML_Comm::sum_all

(

const T &

data

)

References GetType, and m_comm.

Referenced by test_sum_all().

Here is the caller graph for this function:

Member Data Documentation

m_comm

MPI_Comm TML_Comm::m_comm

protected

Referenced by barrier(), broadcast(), broadcast_array(), broadcast_cont_packed(), comm(), exclude(), gather(), gather_debug(), gather_packed(), TML_CartComm::get_coords(), include(), isNull(), operator=(), rank(), receive(), receive_array(), receive_cont_packed(), recv_broadcast(), recv_broadcast_array(), recv_broadcast_cont_packed(), recv_scatter(), recv_scatter_packed(), scatter(), scatter_packed(), send(), send_array(), send_cont_packed(), send_gather(), send_gather_debug(), send_gather_packed(), sendrecv(), sendrecv_array(), sendrecv_cont_packed(), sendrecv_cont_packed_replace(), setComm(), TML_CartComm::shift(), TML_CartComm::shift_array(), TML_CartComm::shift_array_packed(), TML_CartComm::shift_cont(), TML_CartComm::shift_cont_packed(), TML_CartComm::shift_packed(), size(), sum_all(), TML_CartComm::TML_CartComm(), TML_Comm(), and TML_CommWorld::TML_CommWorld().

m_status

MPI_Status TML_Comm::m_status

protected

Referenced by receive(), receive_array(), sendrecv(), and sendrecv_array().

---

The documentation for this class was generated from the following files:

• tml/comm/comm.h
• tml/comm/comm.cpp
• tml/comm/comm.hpp
• tml/comm/comm_coll.hpp