A NetCDF Hyperslab for MPINcFile. More...

Public Member Functions
	MPINcHyperslab (size_t local_start, size_t local_count, MPI_Comm comm)
	`{local_start}, {local_count}`

	MPINcHyperslab (std::vector< size_t > local_start, std::vector< size_t > local_count, MPI_Comm comm)
	`local_start, local_count, comm`

template<class ContainerType , std::enable_if_t< dg::is_vector_v< ContainerType, dg::MPIVectorTag >, bool > = true>
	MPINcHyperslab (const ContainerType &data)
	`{local_start(data) , local_size(data), data.communicator()}`

template<class MPITopology , std::enable_if_t<!dg::is_vector_v< MPITopology >, bool > = true>
	MPINcHyperslab (const MPITopology &grid)
	`grid.start(), grid.count(), grid.communicator()`

template<class T >
	MPINcHyperslab (size_t start0, const T &param)
	Same as `MPINcHyperslab{ start0, 1, grid}`

template<class T >
	MPINcHyperslab (size_t start0, size_t count0, const T &param)
	`{start0, MPINcHyperslab( param).start()}, {count0, MPINcHyperslab(param).count(), MPINcHyperslab.communicator()}`

unsigned	ndim () const

const std::vector< size_t > &	start () const

const std::vector< size_t > &	count () const

std::vector< size_t > &	start ()

std::vector< size_t > &	count ()

MPI_Comm	communicator () const

const size_t *	startp () const

const size_t *	countp () const

Detailed Description

A NetCDF Hyperslab for MPINcFile.

In MPI the data of arrays is usually distributed among processes and each process needs to know where their chunk of data needs to be written in the global array. It is also possible that fewer ranks than are present in file.communicator() actually hold relevant data.

This is how to

See also: specify a hyperslab

Attention: When writing variables, NetCDF-C always assumes that the last dimension of the NetCDF variable varies fastest in the given array. This is in contrast to the default behaviour of our dg::evaluate function, which produces vectors where the first dimension of the given grid varies fastest. Thus, when defining variable dimensions the dimension name of the first grid dimension needs to come last.

Note: The unlimited dimension, if present, must be the first dimension.

Constructor & Destructor Documentation

◆ MPINcHyperslab() [1/6]

dg::file::MPINcHyperslab::MPINcHyperslab	(	size_t	local_start,
		size_t	local_count,
		MPI_Comm	comm )

inline

{local_start}, {local_count}

One dimensional slab

Parameters

local_start	the starting position of a 1d variable
local_count	the count of a 1d variable
comm	communicator of ranks that hold relevant data

◆ MPINcHyperslab() [2/6]

dg::file::MPINcHyperslab::MPINcHyperslab	(	std::vector< size_t >	local_start,
		std::vector< size_t >	local_count,
		MPI_Comm	comm )

inline

local_start, local_count, comm

local_start.size() dimensional slab

Parameters

local_start	specific local start vector
local_count	specific local count vector (must have same size as `local_start`)
comm	communicator of ranks that hold relevant data

◆ MPINcHyperslab() [3/6]

template<class ContainerType , std::enable_if_t< dg::is_vector_v< ContainerType, dg::MPIVectorTag >, bool > = true>

dg::file::MPINcHyperslab::MPINcHyperslab ( const ContainerType & data )

inline

{local_start(data) , local_size(data), data.communicator()}

Infer the local start and count by the size of the data vector. The local size is communicated to all processes in data.communicator() and using the rank one can infer the starting position of the local data chunk. This assumes that the data is ordered by rank.

Attention: This only works for one-dimensional data

Template Parameters

ContainerType ContainerType::size() and ContainerType::communicator() must be callable

Parameters

data	explicitly set one dimensional start and count

Here is the call graph for this function:

◆ MPINcHyperslab() [4/6]

template<class MPITopology , std::enable_if_t<!dg::is_vector_v< MPITopology >, bool > = true>

dg::file::MPINcHyperslab::MPINcHyperslab ( const MPITopology & grid )

inline

grid.start(), grid.count(), grid.communicator()

Template Parameters

MPITopology MPITopolgy::start() and *count() need to return an iterable that can be used to construct std::vector<size_t> MPITopology.communicator() needs to return the communicator of ranks that hold data

Parameters

grid	explicitly set start and count and comm

Attention: When writing variables, NetCDF-C always assumes that the last dimension of the NetCDF variable varies fastest in the given array. This is in contrast to the default behaviour of our dg::evaluate function, which produces vectors where the first dimension of the given grid varies fastest. Thus, when defining variable dimensions the dimension name of the first grid dimension needs to come last.

Note: The unlimited dimension, if present, must be the first dimension.

◆ MPINcHyperslab() [5/6]

template<class T >

dg::file::MPINcHyperslab::MPINcHyperslab	(	size_t	start0,
		const T &	param )

inline

Same as MPINcHyperslab{ start0, 1, grid}

◆ MPINcHyperslab() [6/6]

template<class T >

dg::file::MPINcHyperslab::MPINcHyperslab	(	size_t	start0,
		size_t	count0,
		const T &	param )

inline

{start0, MPINcHyperslab( param).start()}, {count0, MPINcHyperslab(param).count(), MPINcHyperslab.communicator()}

Template Parameters

MPITopology MPITopolgy::start() and *count() need to return an iterable that can be used to construct std::vector<size_t> MPITopology.communicator() needs to return the communicator of ranks that hold data

Parameters

start0	The start coordinate of the unlimited dimension is prepended to the grid.start()
count0	The count coordinate of the unlimited dimension is prepended to the grid.count()
param	explicitly set start and count and comm

Here is the call graph for this function:

Member Function Documentation

◆ communicator()

MPI_Comm dg::file::MPINcHyperslab::communicator ( ) const

inline

Returns: MPI Communicator specifying ranks that participate in reading/writing data

◆ count() [1/2]

std::vector< size_t > & dg::file::MPINcHyperslab::count ( )

inline

Returns: count vector

Here is the call graph for this function:

◆ count() [2/2]

const std::vector< size_t > & dg::file::MPINcHyperslab::count ( ) const

inline

Returns: count vector

Here is the call graph for this function:

◆ countp()

const size_t * dg::file::MPINcHyperslab::countp ( ) const

inline

Returns: pointer to first element of count

Here is the call graph for this function:

◆ ndim()

unsigned dg::file::MPINcHyperslab::ndim ( ) const

inline

Returns: Size of start and count vectors

Here is the call graph for this function:

◆ start() [1/2]

std::vector< size_t > & dg::file::MPINcHyperslab::start ( )

inline

Returns: start vector

Here is the call graph for this function:

◆ start() [2/2]

const std::vector< size_t > & dg::file::MPINcHyperslab::start ( ) const

inline

Returns: start vector

Here is the call graph for this function:

◆ startp()

const size_t * dg::file::MPINcHyperslab::startp ( ) const

inline

Returns: pointer to first element of start

Here is the call graph for this function:

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

nc_hyperslab.h

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ MPINcHyperslab() [1/6]

◆ MPINcHyperslab() [2/6]

◆ MPINcHyperslab() [3/6]

◆ MPINcHyperslab() [4/6]

◆ MPINcHyperslab() [5/6]

◆ MPINcHyperslab() [6/6]

Member Function Documentation

◆ communicator()

◆ count() [1/2]

◆ count() [2/2]

◆ countp()

◆ ndim()

◆ start() [1/2]

◆ start() [2/2]

◆ startp()