Algorithms

Detailed Description

The following algorithms are included in DIY:

Sort

A parallel sample sort that sorts values of each block and computes the boundaries in samples. Optionally, the results can be all_to_all distributed to everyone. A shorter version of the API with a default comparison function and the all_to_all exchange mandatory is also included. The sort uses the same k-ary global reduction as in the communication module.

See the example here.

Kd-tree

A set of points initially distributed in a set of blocks can be redistributed in a kd-tree containing the same number of blocks. The number of blocks must be a power of 2, and the depth of the kd-tree will be log_2(num_blocks). The block boundaries are computed using a histogram; hence they are approximate to the bin width of the histogram. The number of histogram bins is user-supplied.

See the example here.

Functions
template<class Block , class T , class Cmp >
void	diy::sort (Master &master, const Assigner &assigner, std::vector< T > Block::*values, std::vector< T > Block::*samples, size_t num_samples, const Cmp &cmp, int k=2, bool samples_only=false)
	sample sort values of each block, store the boundaries between blocks in samples More...
template<class Block , class T >
void	diy::sort (Master &master, const Assigner &assigner, std::vector< T > Block::*values, std::vector< T > Block::*samples, size_t num_samples, int k=2)
	sample sort values of each block, store the boundaries between blocks in samples shorter version of above sort algorithm with the default less-than comparator used for T and all_to_all exchange included More...
template<class Block , class Point >
void	diy::kdtree (Master &master, const Assigner &assigner, int dim, const ContinuousBounds &domain, std::vector< Point > Block::*points, size_t bins, bool wrap=false)
	build a kd-tree and sort a set of points into it (use histograms to determine split values) More...
template<class Block , class Point >
void	diy::kdtree_sampling (Master &master, const Assigner &assigner, int dim, const ContinuousBounds &domain, std::vector< Point > Block::*points, size_t samples, bool wrap=false)
	build a kd-tree and sort a set of points into it (use sampling to determine split values) More...

Function Documentation

template<class Block , class Point >

void diy::kdtree	(	Master &	master,
		const Assigner &	assigner,
		int	dim,
		const ContinuousBounds &	domain,
		std::vector< Point > Block::*	points,
		size_t	bins,
		bool	wrap = false
	)

build a kd-tree and sort a set of points into it (use histograms to determine split values)

Parameters

master	master object
assigner	assigner object
dim	dimensionality
domain	global data extents
points	input points to sort into kd-tree
bins	number of histogram bins for splitting a dimension
wrap	periodic boundaries in all dimensions

Examples:

reduce/kd-tree.cpp.

template<class Block , class Point >

void diy::kdtree_sampling	(	Master &	master,
		const Assigner &	assigner,
		int	dim,
		const ContinuousBounds &	domain,
		std::vector< Point > Block::*	points,
		size_t	samples,
		bool	wrap = false
	)

build a kd-tree and sort a set of points into it (use sampling to determine split values)

Parameters

master	master object
assigner	assigner object
dim	dimensionality
domain	global data extents
points	input points to sort into kd-tree
samples	number of samples to take in each block
wrap	periodic boundaries in all dimensions

Examples:

reduce/kd-tree.cpp.

template<class Block , class T , class Cmp >

void diy::sort	(	Master &	master,
		const Assigner &	assigner,
		std::vector< T > Block::*	values,
		std::vector< T > Block::*	samples,
		size_t	num_samples,
		const Cmp &	cmp,
		int	k = 2,
		bool	samples_only = false
	)

sample sort values of each block, store the boundaries between blocks in samples

Parameters

master	master object
assigner	assigner object
values	all values to sort
samples	(output) boundaries of blocks
num_samples	desired number of samples
cmp	comparison function
k	k-ary reduction will be used
samples_only	false: results will be all_to_all exchanged; true: only sort but don't exchange results

Examples:

reduce/sample-sort.cpp, and reduce/sort.cpp.

template<class Block , class T >

void diy::sort	(	Master &	master,
		const Assigner &	assigner,
		std::vector< T > Block::*	values,
		std::vector< T > Block::*	samples,
		size_t	num_samples,
		int	k = 2
	)

sample sort values of each block, store the boundaries between blocks in samples shorter version of above sort algorithm with the default less-than comparator used for T and all_to_all exchange included

Parameters

master	master object
assigner	assigner object
values	all values to sort
samples	(output) boundaries of blocks
num_samples	desired number of samples
k	k-ary reduction will be used