util::AggregMax2Dto1D< T, Pr, row > Class Template Reference

Reduces a 2D matrix with stripes into 1D dimension by returning the load of the most loaded stripe. More...

#include <util/prefix_sum_tools.hpp>

---

Detailed Description

template<typename T, typename Pr, bool row>
class util::AggregMax2Dto1D< T, Pr, row >

Reduces a 2D matrix with stripes into 1D dimension by returning the load of the most loaded stripe.

Reduces a 2D matrix with stripes into 1D dimension. Those stripes are created by partitioning in one dimension. Given any interval on the other dimension, returns the load of the most loaded stripe for that interval.

This provides a 1d prefix sum array using the interval notation as defined by util::PrefixSumArray1DOfTInterval.

Parameters:

row

select the orientation of the summation. (e.g. If row is true and the rectangle is of size 10x20, then the equivalent 1darray is of size 10)

List of all members.

Public Member Functions
	AggregMax2Dto1D (const AggregMax2Dto1D &a)
	~AggregMax2Dto1D ()
	AggregMax2Dto1D (const Pr &psum, int procY, int *ycuts)
T	interval (int left, int right) const
	Returns the maximum partition in a given interval. (TODO: please document if this is ]i1:i2] or [i1:i2] or ]i1:i2[ or [i1:i2[).
int	getLineCount ()
Private Member Functions
	AggregMax2Dto1D ()
const AggregMax2Dto1D &	operator= (const AggregMax2Dto1D &a)
Private Attributes
int	lineCount
	number of fixed rows or columns
Aggreg2Dto1D< T, Pr, row > **	lineSet

---

Constructor & Destructor Documentation

template<typename T, typename Pr, bool row>

util::AggregMax2Dto1D< T, Pr, row >::AggregMax2Dto1D

(

)

[private]

template<typename T , typename Pr , bool row>

util::AggregMax2Dto1D< T, Pr, row >::AggregMax2Dto1D

(

const AggregMax2Dto1D< T, Pr, row > &

a

)

Copy constructor for AggregMax2Dto1D

Parameters:

a

template<typename T , typename Pr , bool row>

util::AggregMax2Dto1D< T, Pr, row >::~AggregMax2Dto1D

(

)

template<typename T , typename Pr , bool row>

util::AggregMax2Dto1D< T, Pr, row >::AggregMax2Dto1D	(	const Pr &	psum,
		int	procY,
		int *	ycuts
	)

Constructor

Parameters:

	psum	2D prefix sum Matrix
	procY	number of processors in X or Y dimension (depending on orientation)
	ycuts	fixed cut points for X or Y dimension (TODO: format ?)

---

Member Function Documentation

template<typename T , typename Pr , bool row>

int util::AggregMax2Dto1D< T, Pr, row >::getLineCount

(

)

AggregMax2Dto1D is built by defining a number of stripes in the first dimension. This method returns the number of stripes (chunk of rows/columns in the matrix).

Returns:

number of stripes

template<typename T , typename Pr , bool row>

T util::AggregMax2Dto1D< T, Pr, row >::interval	(	int	left,
		int	right
	)			const

Returns the maximum partition in a given interval. (TODO: please document if this is ]i1:i2] or [i1:i2] or ]i1:i2[ or [i1:i2[).

Parameters:

	left	lower interval bound
	right	higher interval bound

Returns:

max partition weight in the given interval

template<typename T, typename Pr, bool row>

const AggregMax2Dto1D& util::AggregMax2Dto1D< T, Pr, row >::operator=

(

const AggregMax2Dto1D< T, Pr, row > &

a

)

[private]

---

Member Data Documentation

template<typename T, typename Pr, bool row>

int util::AggregMax2Dto1D< T, Pr, row >::lineCount [private]

number of fixed rows or columns

template<typename T, typename Pr, bool row>

Aggreg2Dto1D<T,Pr,row>** util::AggregMax2Dto1D< T, Pr, row >::lineSet [private]

chunks of partitioned lines (can be row-wise or column-wise). For example: when we parition row-wise, we get line chunks like chunk[0]={l1,l2,l3} chunk[1]={l4} etc.

---

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

• include/util/prefix_sum_tools.hpp
• include/util/prefix_sum_tools_impl.hpp