#include <server_pack.h>

Inheritance diagram for fpmas::synchro::hard::ServerPackBase:

[legend]

Collaboration diagram for fpmas::synchro::hard::ServerPackBase:

Public Member Functions
	ServerPackBase (fpmas::api::communication::MpiCommunicator &comm, api::TerminationAlgorithm &termination, api::Server &mutex_server, api::Server &link_server)

virtual api::Server &	mutexServer ()

virtual api::Server &	linkServer ()

void	handleIncomingRequests () override

void	setEpoch (Epoch epoch) override

Epoch	getEpoch () const override

void	terminate ()

std::vector< fpmas::api::communication::Request > &	pendingRequests ()

void	waitSendRequest (fpmas::api::communication::Request &req)

template<typename T >
void	waitResponse (fpmas::api::communication::TypedMpi< T > &mpi, int source, api::Tag tag, fpmas::api::communication::Status &status)

void	waitVoidResponse (fpmas::api::communication::MpiCommunicator &comm, int source, api::Tag tag, fpmas::api::communication::Status &status)

virtual void	setEpoch (Epoch epoch)=0

virtual Epoch	getEpoch () const =0

virtual void	handleIncomingRequests ()=0

Detailed Description

A Server implementation wrapping an api::MutexServer and an api::LinkServer.

This is used to apply an api::TerminationAlgorithm to the two instances simultaneously. More precisely, when applying the termination algorithm to this server, the system is still able to answer to mutex AND link requests, to completely determine global termination.

The wait*() methods also allows to easily wait for point-to-point communications initiated by either server without deadlock, by ensuring progression on both servers.

Constructor & Destructor Documentation

◆ ServerPackBase()

fpmas::synchro::hard::ServerPackBase::ServerPackBase	(	fpmas::api::communication::MpiCommunicator &	comm,
		api::TerminationAlgorithm &	termination,
		api::Server &	mutex_server,
		api::Server &	link_server
	)

inline

ServerPackBase constructor.

Parameters

comm	MPI communicator
termination	termination algorithm used to terminated this server, terminating both mutex_server and link_server at once.
mutex_server	Mutex server
link_server	Link server

Member Function Documentation

◆ mutexServer()

virtual api::Server & fpmas::synchro::hard::ServerPackBase::mutexServer ( )

inlinevirtual

Reference to the internal api::MutexServer instance.

Reimplemented in fpmas::synchro::hard::ServerPack< MutexServer, LinkServer >, fpmas::synchro::hard::ServerPack< fpmas::synchro::hard::api::MutexServer< T >, fpmas::synchro::hard::VoidServer >, and fpmas::synchro::hard::ServerPack< fpmas::synchro::hard::api::MutexServer< T >, fpmas::synchro::hard::api::LinkServer >.

◆ linkServer()

virtual api::Server & fpmas::synchro::hard::ServerPackBase::linkServer ( )

inlinevirtual

Reference to the internal api::LinkServer instance.

Reimplemented in fpmas::synchro::hard::ServerPack< MutexServer, LinkServer >, fpmas::synchro::hard::ServerPack< fpmas::synchro::hard::api::MutexServer< T >, fpmas::synchro::hard::VoidServer >, and fpmas::synchro::hard::ServerPack< fpmas::synchro::hard::api::MutexServer< T >, fpmas::synchro::hard::api::LinkServer >.

◆ handleIncomingRequests()

void fpmas::synchro::hard::ServerPackBase::handleIncomingRequests ( )

inlineoverridevirtual

Performs a reception cycle on the api::MutexServer AND on the api::LinkServer.

Implements fpmas::synchro::hard::api::Server.

◆ setEpoch()

void fpmas::synchro::hard::ServerPackBase::setEpoch ( Epoch epoch )

inlineoverridevirtual

Sets the Epoch of the api::MutexServer AND the api::LinkServer.

Parameters

epoch new epoch

Implements fpmas::synchro::hard::api::Server.

◆ getEpoch()

Epoch fpmas::synchro::hard::ServerPackBase::getEpoch ( ) const

inlineoverridevirtual

Gets the common epoch of the api::MutexServer and the api::LinkServer.

Returns: current Epoch

Implements fpmas::synchro::hard::api::Server.

◆ terminate()

void fpmas::synchro::hard::ServerPackBase::terminate ( )

inline

Applies the termination algorithm to this ServerPack.

◆ pendingRequests()

std::vector< fpmas::api::communication::Request > & fpmas::synchro::hard::ServerPackBase::pendingRequests ( )

inline

Returns a list containing all the pending non-blocking requests, that are waiting for completions.

Such non-blocking communications are notably used to send responses to READ and ACQUIRE requests.

Pending requests are guaranteed to be completed at the latest upon return of the next terminate() call, so that Request buffers can be freed.

It is also valid to complete requests before the next terminate() call, for example using fpmas::api::communication::MpiCommunicator::test() or fpmas::api::communication::MpiCommunicator::testSome() in order to limit memory usage, as long as this cannot produce deadlock situations.

◆ waitSendRequest()

void fpmas::synchro::hard::ServerPackBase::waitSendRequest ( fpmas::api::communication::Request & req )

inline

Method used to wait for an MPI_Request to be sent.

The request might be performed by the api::MutexServer OR the api::LinkServer. In any case, this methods handles incoming requests (see api::MutexServer::handleIncomingRequests() and api::LinkServer::handleIncomingRequests()) until the MPI_Request is complete, in order to avoid deadlock.

Parameters

req	MPI request to complete

◆ waitResponse()

template<typename T >

void fpmas::synchro::hard::ServerPackBase::waitResponse	(	fpmas::api::communication::TypedMpi< T > &	mpi,
		int	source,
		api::Tag	tag,
		fpmas::api::communication::Status &	status
	)

inline

Method used to wait for request responses with data (e.g. to READ, ACQUIRE) from the specified mpi instance to be available.

The request might be performed by the api::MutexServer OR the api::LinkServer. In any case, this methods handles incoming requests (see api::MutexServer::handleIncomingRequests() and api::LinkServer::handleIncomingRequests()) until a response is available, in order to avoid deadlock.

Upon return, it is safe to receive the response with api::communication::MpiCommunicator::recv() using the same source and tag.

Parameters

mpi	TypedMpi instance to probe
source	rank of the process from which the response should be received
tag	response tag (READ_RESPONSE, ACQUIRE_RESPONSE, etc)
status	pointer to MPI_Status, passed to api::communication::MpiCommunicator::Iprobe

◆ waitVoidResponse()

void fpmas::synchro::hard::ServerPackBase::waitVoidResponse	(	fpmas::api::communication::MpiCommunicator &	comm,
		int	source,
		api::Tag	tag,
		fpmas::api::communication::Status &	status
	)

inline

Method used to wait for request responses without data (e.g. to LOCK, UNLOCK) from the specified comm instance to be available.

The request might be performed by the api::MutexServer OR the api::LinkServer. In any case, this methods handles incoming requests (see api::MutexServer::handleIncomingRequests() and api::LinkServer::handleIncomingRequests()) until a response is available, in order to avoid deadlock.

Upon return, it is safe to receive the response with api::communication::MpiCommunicator::recv() using the same source and tag.

Parameters

comm	MpiCommunicator instance to probe
source	rank of the process from which the response should be received
tag	response tag (READ_RESPONSE, ACQUIRE_RESPONSE, etc)
status	pointer to MPI_Status, passed to api::communication::MpiCommunicator::Iprobe

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

src/fpmas/synchro/hard/server_pack.h

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ ServerPackBase()

Member Function Documentation

◆ mutexServer()

◆ linkServer()

◆ handleIncomingRequests()

◆ setEpoch()

◆ getEpoch()

◆ terminate()

◆ pendingRequests()

◆ waitSendRequest()

◆ waitResponse()

◆ waitVoidResponse()