MPI info. More...

#include <parallel.h>

Public Member Functions
	Parallel (int argc, char **argv, bool active, int groupsize)
	Constructor. More...

	~Parallel ()

bool	IamMaster () const
	Returns true if this process is the master process. More...

bool	IamGroupMaster () const
	Returns true if this process is the master process within a group. More...

bool	isMyWork (int i) const
	Returns true if the work item is assigned to this process. More...

bool	isMyGroupWork (int i) const
	Returns true if the work item is assigned to this process' group. More...

bool	active () const
	Returns true if the MPI is active. More...

int	iproc () const
	Returns the rank of the communicator (process is). More...

int	igroup () const
	Returns the rank of process group. More...

int	igroupproc () const
	Returns index of rank within the group. More...

int	Nproc () const
	Returns the size of the communicator (process count). More...

int	Ngroup () const
	Returns the number of groups. More...

int	groupsize () const
	Return group size. More...

template<class T >
void	bcast (int owner, T *data, std::size_t N) const
	Broadcast array from owner to everyone. More...

template<class T >
void	bcast (int owner, NumberArray< T > &data) const
	Broadcast array from owner to everyone. More...

template<class T >
void	bcast_g (int igroup, int groupowner, T *data, std::size_t N) const
	Broadcast array from owner to everyone in the group. More...

template<class T >
void	send (T const *array, std::size_t size, int origin, int destination) const
	Send data to a process. More...

template<class T >
void	recv (T *array, std::size_t size, int origin, int destination) const
	Receive data from a process. More...

template<class T >
void	sync (T *array, std::size_t chunksize, std::size_t Nchunk) const
	Synchronize across processes by composition. More...

template<class T >
void	sync_m (T *array, std::size_t chunksize, std::size_t Nchunk) const

template<class T >
void	sum (T *array, std::size_t N, int owner=0) const
	Sum arrays to node. More...

template<class T >
void	syncsum (T *array, std::size_t N) const
	Synchronize across processes by summing. More...

template<class T >
void	syncsum_g (T *array, std::size_t N) const
	Synchronize across group's processes by summing. More...

template<class T >
void	sum_g (T *array, std::size_t N, int destination) const
	Sum array within group to a given process. More...

template<class T >
void	mastersum (T *array, std::size_t N, int destgroup) const
	Sum array from all group masters to one of the group masters. More...

void	wait () const
	Wait for completition of all running tasks. More...

void	wait_g () const

void *	groupcomm () const

Detailed Description

The class Parallel holds some useful MPI data, like the rank and size of the communicator. This class should be used for all parallel communication using MPI.

Constructor & Destructor Documentation

◆ Parallel()

Parallel::Parallel	(	int *	argc,
		char ***	argv,
		bool	active,
		int	groupsize
	)

inline

Initializes MPI (using the MPI_Init) function. Also determines the rank and size of the communicator.

Parameters

argc	Argc as received by main().
argv	Argv as received by main().
active	Whether to turn the MPI on or keep with the computation sequential.

◆ ~Parallel()

Parallel::~Parallel ( )

inline

Member Function Documentation

◆ active()

bool Parallel::active ( ) const

inline

◆ bcast() [1/2]

template<class T >

void Parallel::bcast	(	int	owner,
		T *	data,
		std::size_t	N
	)		const

inline

◆ bcast() [2/2]

template<class T >

void Parallel::bcast	(	int	owner,
		NumberArray< T > &	data
	)		const

inline

◆ bcast_g()

template<class T >

void Parallel::bcast_g	(	int	igroup,
		int	groupowner,
		T *	data,
		std::size_t	N
	)		const

inline

◆ groupcomm()

void* Parallel::groupcomm ( ) const

inline

◆ groupsize()

int Parallel::groupsize ( ) const

inline

◆ IamGroupMaster()

bool Parallel::IamGroupMaster ( ) const

inline

This function returns true if this process is the master process within its, i.e. it has local ID zero.

◆ IamMaster()

bool Parallel::IamMaster ( ) const

inline

This function returns true if this process is the master process, i.e. it has ID zero.

◆ igroup()

int Parallel::igroup ( ) const

inline

◆ igroupproc()

int Parallel::igroupproc ( ) const

inline

◆ iproc()

int Parallel::iproc ( ) const

inline

◆ isMyGroupWork()

bool Parallel::isMyGroupWork ( int i ) const

inline

If there are several work items, i-th one is assigned to the (i % Ngroup)-th group. This function returns true whenever given work item (indexed by "i") is to be processed by the current process' group.

Parameters

i	Work item ID.

◆ isMyWork()

bool Parallel::isMyWork ( int i ) const

inline

If there are several work items, i-th one is assigned to the (i % Nproc_)-th process. This function returns true whenever given work item (indexed by "i") is to be processed by the current process.

Parameters

i	Work item ID.

◆ mastersum()

template<class T >

void Parallel::mastersum	(	T *	array,
		std::size_t	N,
		int	destgroup
	)		const

inline

◆ Ngroup()

int Parallel::Ngroup ( ) const

inline

◆ Nproc()

int Parallel::Nproc ( ) const

inline

◆ recv()

template<class T >

void Parallel::recv	(	T *	array,
		std::size_t	size,
		int	origin,
		int	destination
	)		const

inline

◆ send()

template<class T >

void Parallel::send	(	T const *	array,
		std::size_t	size,
		int	origin,
		int	destination
	)		const

inline

◆ sum()

template<class T >

void Parallel::sum	(	T *	array,
		std::size_t	N,
		int	owner = `0`
	)		const

inline

◆ sum_g()

template<class T >

void Parallel::sum_g	(	T *	array,
		std::size_t	N,
		int	destination
	)		const

inline

◆ sync()

template<class T >

void Parallel::sync	(	T *	array,
		std::size_t	chunksize,
		std::size_t	Nchunk
	)		const

inline

Synchronize array across processes. It is assumed that i-th chunk of the array is present on the (i % Nproc_)-th process. That process will be used as the broadcast root. This behaviour is compatible with the member function isMyWork.

Parameters

array	Pointer to data array to synchronize.
chunksize	Size of the per-process segment.
Nchunk	Total number of chunks in the array. Altogether chunksize*Nchunk elements will be synchronized. If there are some elements more, they will be left untouched (and un-broadcast).

It is expected that the array has length equal or greater than chunksize * Nchunk.

◆ sync_m()

template<class T >

void Parallel::sync_m	(	T *	array,
		std::size_t	chunksize,
		std::size_t	Nchunk
	)		const

inline

◆ syncsum()

template<class T >

void Parallel::syncsum	(	T *	array,
		std::size_t	N
	)		const

inline

Synchronize array across processes by summing.

Parameters

array	Pointer to data array to synchronize.
Nchunk	Number of elements in the array to sum-synchronize. If there are some elements more, they will be left untouched (and un-broadcast).

◆ syncsum_g()

template<class T >

void Parallel::syncsum_g	(	T *	array,
		std::size_t	N
	)		const

inline

Synchronize array across group's processes by summing.

Parameters

array	Pointer to data array to synchronize.
Nchunk	Number of elements in the array to sum-synchronize. If there are some elements more, they will be left untouched (and un-broadcast).

◆ wait()

void Parallel::wait ( ) const

inline

Inserts a MPI BARRIER.

◆ wait_g()

void Parallel::wait_g ( ) const

inline

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

src/parallel.h

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ Parallel()

◆ ~Parallel()

Member Function Documentation

◆ active()

◆ bcast() [1/2]

◆ bcast() [2/2]

◆ bcast_g()

◆ groupcomm()

◆ groupsize()

◆ IamGroupMaster()

◆ IamMaster()

◆ igroup()

◆ igroupproc()

◆ iproc()

◆ isMyGroupWork()

◆ isMyWork()

◆ mastersum()

◆ Ngroup()

◆ Nproc()

◆ recv()

◆ send()

◆ sum()

◆ sum_g()

◆ sync()

◆ sync_m()

◆ syncsum()

◆ syncsum_g()

◆ wait()

◆ wait_g()