Files for communication of parallel solvers. More...

Description

Classes
class	Dumux::ScotchPartitioner< IndexType >
	A reordering backend using the scotch library. More...
class	Dumux::VectorCommDataHandle< Mapper, Vector, entityCodim, ScatterOperator, DataT >
	A data handle class to exchange entries of a vector. More...
class	Dumux::MultiCodimVectorCommDataHandle< Mapper, Vector, dim, ScatterOperator, DataT >
	A data handle class to exchange entries of a vector for multiple codims in one communication call. More...

Functions
constexpr bool	Dumux::Multithreading::isSerial ()
	Checking whether the backend is serial.
template<class Communication>
std::vector< char >	Dumux::Detail::exchangeSparse (const Communication &comm, std::vector< std::vector< char > > &sendBufs, int tag=7373)
	Exchange variable-sized byte messages with a sparse set of peer processes.
template<class FunctorType>
void	Dumux::parallelFor (const std::size_t count, const FunctorType &functor)
	A parallel for loop (multithreading).

Files
file	multithreading.hh
	Multithreading in Dumux.
file	nonblockingsparseexchange.hh
	A dynamic sparse data exchange (DSDE) of byte messages between processes.
file	packing.hh
	Helpers to (de)serialize trivially-copyable values into a byte buffer (e.g. for point-to-point or collective MPI communication).
file	parallel_for.hh
	Parallel for loop (multithreading).
file	scotchpartitioner.hh
	An interface to the Scotch library for graph partitioning.
file	vectorcommdatahandle.hh
	Contains a class to exchange entries of a vector.

Function Documentation

template<class Communication>

std::vector< char > Dumux::Detail::exchangeSparse	(	const Communication &	comm,
		std::vector< std::vector< char > > &	sendBufs,
		int	tag = 7373 )

Parameters

comm	a Dune communicator (convertible to an MPI_Comm when built with MPI)
sendBufs	one byte buffer per destination rank; an empty buffer means "no message for that rank". A buffer addressed to the calling rank is ignored.
tag	the MPI message tag to use; it must not clash with other communication in flight on the same communicator while this exchange runs.

Returns: the concatenation, in unspecified order, of all messages addressed to this rank

Note: Collective over comm. Each rank only needs to know its own destinations; who sends to it (and how much) is discovered on the fly, so there is no O(number of ranks) memory or all-to-all collective involved.

bool Dumux::Multithreading::isSerial ( )

inlineconstexpr

template<class FunctorType>

void Dumux::parallelFor	(	const std::size_t	count,
		const FunctorType &	functor )

inline

Parameters

count	the number of work tasks to perform
functor	functor executed for each task (get task number as argument)