A sparse matrix container class that mainly stores its data in CSR format. More...

#include <sparse_matrix.h>

Classes
class	iterator
	An sparse matrix iterator class that tranverses row-wise. More...

Public Member Functions
	SparseMatrix (size_t rows, size_t cols, size_t nnz, T values, size_t rowPtr, size_t *colInd, bool dealloc=true, T zeroValue=T(), bool transMatrix=false)

	SparseMatrix (size_t rows, size_t cols, size_t nnz, T min, T max, T zeroValue=T())

	SparseMatrix (const SparseMatrix &copy)

	SparseMatrix (const std::string &inputfile, enum SparseFileFormat format=MATRIX_MARKET_FORMAT, T zeroValue=T())

	~SparseMatrix ()

void	operator= (const SparseMatrix< T > &copy)

size_t	total_nnz () const

size_t	total_rows () const

size_t	total_cols () const

T *	get_values () const

iterator	begin (unsigned row)

T	at (size_t row, size_t col) const

const T &	at (size_t index) const

const T &	operator() (const size_t row, const size_t col) const

T	operator() (const size_t row, const size_t col)

void	resize (SparseMatrix< T > &copy, bool retainData)

void	updateHost () const

void	invalidateDevice ()

void	updateHostAndInvalidateDevice ()

device_pointer_type_cl	updateDevice_CL (T start, size_t elems, Device_CL device, bool copy)
	Update device with sparse matrix content. More...

device_pointer_index_type_cl	updateDevice_Index_CL (size_t start, size_t elems, Device_CL device, bool copy)
	Update device with sparse matrix index contents that have an "size_t" type. More...

void	flush_CL ()
	Flushes the sparse matrix. More...

device_pointer_type_cu	updateDevice_CU (T *start, size_t elems, unsigned int deviceID, bool copy)
	Update device with matrix content. More...

device_pointer_index_type_cu	updateDevice_Index_CU (size_t *start, size_t elems, unsigned int deviceID, bool copy)
	Update device with sparse matrix index contents that have an "size_t" type. More...

void	flush_CU ()
	Flushes the matrix. More...

bool	isSparseMatrixOnDevice_CU (unsigned int deviceID)

bool	isModified_CU (unsigned int deviceID)

Friends
std::ostream &	operator<< (std::ostream &os, SparseMatrix< T > &matrix)
	Overloaded stream operator, for testing purposes. More...

Detailed Description

template<class T>
class skepu::SparseMatrix< T >

A sparse matrix container class that mainly stores its data in CSR format.

A skepu::SparseMatrix is a container for storing sparse 2D structures that are internally stores in a 1D array using CSR format. It supports operation to load sparse matrices from file as well as creating random sparse matrices. As CSR format, it stores 3 arrays: actual values, column indices and row offsets. It also keeps track of which parts of it are currently allocated and uploaded to the GPU. If a computation is done, changing the elements in the GPU memory, it is not directly transferred back to the host memory. Instead, the SparseMatrix waits until an element is accessed before any copying is done. The class also supports transpose operation which it achieves by converting CSR (Compressed Storage Rows) format to CSC (Compressed Storage Column) format which can be useful e.g. when applying column-wise reduction operation. The transpose can be obtained by simply '~' operation, e.g., transpose_m0 = ~m0;

It also implements support for allocating and de-allocating page-locked memory using cudaMallocHost and cudaFreeHost. This could help is running asynchronous operations especially when using multiple CUDA devices. It can be enabled by defining USE_PINNED_MEMORY flag in the skeleton program.

Constructor & Destructor Documentation

template<typename T>

skepu::SparseMatrix< T >::SparseMatrix	(	size_t	rows,
		size_t	cols,
		size_t	nnz,
		T *	values,
		size_t *	rowPtr,
		size_t *	colInd,
		bool	dealloc = `true`,
		T	zeroValue = `T()`,
		bool	transMatrix = `false`
	)

SparseMatrix Constructor, used to create a sparse matrix with given data (rows-offsets,cols-indices and non-zero elements).

Parameters

rows	Number of rows in the sparse matrix.
cols	Number of columns in the sparse matrix.
nnz	Number of non-zero elements in the sparse matrix.
values	An array containing non-zero elements stored row-wise in C order.
rowPtr	An array containing indices pointing to starting indices of each row in the values array.
colInd	An array containing indices pointing to column indices for each element in the values array.
dealloc	A boolean defining whether the arrays are going to be de-allocated when the destructor is called.
zeroValue	value that represent zero value for the given elements type, default will be initial value of that data type.
transMatrix	A boolean that specifies whether the matrix is a transpose matrix or a normal one.

template<typename T>

skepu::SparseMatrix< T >::SparseMatrix	(	size_t	rows,
		size_t	cols,
		size_t	nnz,
		T	min,
		T	max,
		T	zeroValue = `T()`
	)

SparseMatrix Constructor, used to generate a random sparse matrix (in privided min max range) with given rows,cols and non-zero elements.

Parameters

rows	Number of rows in the sparse matrix.
cols	Number of columns in the sparse matrix.
nnz	Number of non-zero elements in the sparse matrix.
min	Minimum value for the random values generated
max	Maximum value for the random values generated
zeroValue	value that represent zero value for the given elements type, default will be initial value of that data type.

References skepu::get_random_number().

Here is the call graph for this function:

template<typename T>

skepu::SparseMatrix< T >::SparseMatrix ( const SparseMatrix< T > & copy )

SparseMatrix Copy Constructor, used to create a copy of another sparse matrix. Does not copy the transpose matrix part.

Parameters

copy	sparse matrix which we are aopying from.

template<typename T>

skepu::SparseMatrix< T >::SparseMatrix	(	const std::string &	inputfile,
		enum SparseFileFormat	format = `MATRIX_MARKET_FORMAT`,
		T	zeroValue = `T()`
	)

SparseMatrix Constructor, used to read a sparse matrix from a text file (preferably in MTX format) with given rows,cols and non-zero elements.

Parameters

inputfile	Name of the input file.
format	Format of the matrix storage, currently only MATRIX_MARKET_FORMAT is supported.
zeroValue	value that represent zero value for the given elements type, default will be initial value of that data type.

template<typename T >

skepu::SparseMatrix< T >::~SparseMatrix ( )

SparseMatrix Destructor. Internally Deallocates memory if proper flags are set.

Member Function Documentation

template<typename T >

T skepu::SparseMatrix< T >::at	(	size_t	row,
		size_t	col
	)		const

Uses row and col instead to find element. If found, return actual otherwise returns 0

Parameters

row	Index of row to get.
col	Index of column to get.

template<typename T >

const T & skepu::SparseMatrix< T >::at ( size_t index ) const

Directly access an element by specifying its index. If found, return actual otherwise returns 0

Parameters

index Index of row to get.

template<typename T >

SparseMatrix< T >::iterator skepu::SparseMatrix< T >::begin ( unsigned row )

inline

Please refer to the documentation of std::vector and skepu::Matrix::iterator. Uses row to get an iterator for that row.

Parameters

row	The index of row from where to start iterator.

Referenced by skepu::Reduce< ReduceFunc, ReduceFunc >::CL(), skepu::Reduce< ReduceFuncRowWise, ReduceFuncColWise >::CPU(), skepu::Reduce< ReduceFunc, ReduceFunc >::CPU(), skepu::Reduce< ReduceFunc, ReduceFunc >::CU(), skepu::ompIrregularWorkload(), and skepu::ompRegularWorkload().

template<typename T >

void skepu::SparseMatrix< T >::flush_CL ( )

Flushes the sparse matrix.

First it updates the sparse matrix from all its device allocations, then it releases all allocations.

template<typename T >

void skepu::SparseMatrix< T >::flush_CU ( )

Flushes the matrix.

First it updates the matrix from all its device allocations, then it releases all allocations.

template<class T>

T* skepu::SparseMatrix< T >::get_values ( ) const

inline

Returns pointer to actual non zero values in the SparseMatrix.

Returns: pointer to actual non zero values in the SparseMatrix.

Referenced by skepu::MapArray< MapArrayFunc >::CL(), skepu::Reduce< ReduceFunc, ReduceFunc >::CL(), skepu::Map< MapFunc >::CL(), skepu::MapArray< MapArrayFunc >::CPU(), skepu::Map< MapFunc >::CPU(), skepu::Reduce< ReduceFunc, ReduceFunc >::CPU(), skepu::MapArray< MapArrayFunc >::CU(), skepu::Map< MapFunc >::CU(), skepu::Reduce< ReduceFunc, ReduceFunc >::CU(), skepu::MapArray< MapArrayFunc >::OMP(), skepu::Map< MapFunc >::OMP(), and skepu::Reduce< ReduceFunc, ReduceFunc >::OMP().

template<typename T >

void skepu::SparseMatrix< T >::invalidateDevice ( )

inline

Invalidates all device data that this sparse matrix has allocated.

template<typename T >

bool skepu::SparseMatrix< T >::isModified_CU ( unsigned int deviceID )

Can be used to query whether sparse matrix is modified on a device or not.

Referenced by skepu::Reduce< ReduceFuncRowWise, ReduceFuncColWise >::operator()(), skepu::Map< MapFunc >::operator()(), skepu::MapArray< MapArrayFunc >::operator()(), and skepu::Reduce< ReduceFunc, ReduceFunc >::operator()().

template<typename T >

bool skepu::SparseMatrix< T >::isSparseMatrixOnDevice_CU ( unsigned int deviceID )

Can be used to query whether matrix is already available on a device or not.

Referenced by skepu::Reduce< ReduceFuncRowWise, ReduceFuncColWise >::operator()(), skepu::Map< MapFunc >::operator()(), skepu::MapArray< MapArrayFunc >::operator()(), and skepu::Reduce< ReduceFunc, ReduceFunc >::operator()().

template<typename T >

const T & skepu::SparseMatrix< T >::operator()	(	const size_t	row,
		const size_t	col
	)		const

Can be used to access elements by specifying row and column index. it cares about synchronizing with device. Can be used when accessing to access elements row and column wise.

Parameters

row	Index to a specific row of the sparse matrix.
col	Index to a specific column of the sparse matrix.

template<typename T >

T skepu::SparseMatrix< T >::operator()	(	const size_t	row,
		const size_t	col
	)

Behaves like operator[] and unlike skepu::Vector, it cares about synchronizing with device. Can be used when accessing to access elements row and column wise.

Parameters

row	Index to a specific row of the sparse matrix.
col	Index to a specific column of the sparse matrix.

template<typename T>

void skepu::SparseMatrix< T >::operator= ( const SparseMatrix< T > & other )

SparseMatrix Copy Constructor, used to create a copy of another sparse matrix. Does not copy the transpose matrix part.

Parameters

copy	sparse matrix which we are aopying from.

template<typename T>

void skepu::SparseMatrix< T >::resize	(	SparseMatrix< T > &	copy,
		bool	retainData
	)

will resize the matrix, can be dangerous, used with care

Referenced by skepu::Map< MapFunc >::operator()().

template<class T>

size_t skepu::SparseMatrix< T >::total_cols ( ) const

inline

Returns total number of columns in the SparseMatrix.

Returns: columns in the SparseMatrix.

Referenced by skepu::MapArray< MapArrayFunc >::CL(), skepu::Reduce< ReduceFuncRowWise, ReduceFuncColWise >::CPU(), skepu::MapArray< MapArrayFunc >::CPU(), skepu::Reduce< ReduceFunc, ReduceFunc >::CPU(), skepu::MapArray< MapArrayFunc >::CU(), skepu::MapArray< MapArrayFunc >::OMP(), and skepu::Reduce< ReduceFuncRowWise, ReduceFuncColWise >::operator()().

template<class T>

size_t skepu::SparseMatrix< T >::total_nnz ( ) const

inline

Returns number of non zero elements in the SparseMatrix.

Returns: count of non zero elements of the SparseMatrix.

template<class T>

size_t skepu::SparseMatrix< T >::total_rows ( ) const

inline

Returns total number of rows in the SparseMatrix.

Returns: rows in the SparseMatrix.

Referenced by skepu::MapArray< MapArrayFunc >::CL(), skepu::Reduce< ReduceFunc, ReduceFunc >::CL(), skepu::Reduce< ReduceFuncRowWise, ReduceFuncColWise >::CPU(), skepu::MapArray< MapArrayFunc >::CPU(), skepu::Reduce< ReduceFunc, ReduceFunc >::CPU(), skepu::MapArray< MapArrayFunc >::CU(), skepu::Reduce< ReduceFunc, ReduceFunc >::CU(), skepu::Reduce< ReduceFuncRowWise, ReduceFuncColWise >::OMP(), skepu::MapArray< MapArrayFunc >::OMP(), skepu::Reduce< ReduceFunc, ReduceFunc >::OMP(), skepu::ompIrregularWorkload(), skepu::ompRegularWorkload(), and skepu::Reduce< ReduceFuncRowWise, ReduceFuncColWise >::operator()().

template<typename T>

SparseMatrix< T >::device_pointer_type_cl skepu::SparseMatrix< T >::updateDevice_CL	(	T *	start,
		size_t	elems,
		Device_CL *	device,
		bool	copy
	)

Update device with sparse matrix content.

Update device with a SparseMatrix range by specifying number of elements. If sparse matrix does not have an allocation on the device for the current range, create a new allocation and if specified, also copy sparse matrix data to device. Saves newly allocated ranges to m_deviceMemPointers_CL so sparse matrix can keep track of where and what it has stored on devices.

Parameters

start	Pointer to first element in range to be updated with device.
elems	Number of (non-zero) elements.
device	Pointer to the device that should be synched with.
copy	Boolean value that tells whether to only allocate or also copy sparse matrix data to device. True copies, False only allocates.

References skepu::DeviceMemPointer_CL< T >::copyHostToDevice(), and skepu::Device_CL::getDeviceID().

Referenced by skepu::MapArray< MapArrayFunc >::CL(), skepu::Reduce< ReduceFunc, ReduceFunc >::CL(), and skepu::Map< MapFunc >::CL().

Here is the call graph for this function:

template<typename T>

SparseMatrix< T >::device_pointer_type_cu skepu::SparseMatrix< T >::updateDevice_CU	(	T *	start,
		size_t	elems,
		unsigned int	deviceID,
		bool	copy
	)

Update device with matrix content.

Update device with a SparseMatrix range by specifying array of non-zero elements. If SparseMatrix does not have an allocation on the device for the current range, create a new allocation and if specified, also copy SparseMatrix data to device. Saves newly allocated ranges to m_deviceMemPointers_CU so SparseMatrix can keep track of where and what it has stored on devices.

Parameters

start	Pointer to first (non-zero) element in range to be updated with device.
elems	Number of (non-zero) elements.
deviceID	Integer specififying the device that should be synched with.
copy	boolean value that tells whether to only allocate or also copy sparse matrix data to device. True copies, false only allocates.

References skepu::DeviceMemPointer_CU< T >::copyHostToDevice(), and skepu::Environment< T >::getInstance().

Referenced by skepu::MapArray< MapArrayFunc >::CU(), skepu::Map< MapFunc >::CU(), and skepu::Reduce< ReduceFunc, ReduceFunc >::CU().

Here is the call graph for this function:

template<typename T >

SparseMatrix< T >::device_pointer_index_type_cl skepu::SparseMatrix< T >::updateDevice_Index_CL	(	size_t *	start,
		size_t	elems,
		Device_CL *	device,
		bool	copy
	)

Update device with sparse matrix index contents that have an "size_t" type.

Update device with a SparseMatrix index range by specifying number of elements of "size_t" type. If sparse matrix does not have an allocation on the device for the current range, create a new allocation and if specified, also copy sparse matrix data to device. Saves newly allocated ranges to m_deviceMemIndexPointers_CL so sparse matrix can keep track of where and what it has stored on devices.

Parameters

start	Pointer to first index element in range to be updated with device.
elems	Number of (non-zero) index elements.
device	Pointer to the device that should be synched with.
copy	Boolean value that tells whether to only allocate or also copy sparse matrix index data to device. True copies, False only allocates.

References skepu::DeviceMemPointer_CL< T >::copyHostToDevice(), and skepu::Device_CL::getDeviceID().

Referenced by skepu::MapArray< MapArrayFunc >::CL().

Here is the call graph for this function:

template<typename T >

SparseMatrix< T >::device_pointer_index_type_cu skepu::SparseMatrix< T >::updateDevice_Index_CU	(	size_t *	start,
		size_t	elems,
		unsigned int	deviceID,
		bool	copy
	)

Update device with sparse matrix index contents that have an "size_t" type.

Update device with a SparseMatrix index range by specifying number of elements of "size_t" type. If sparse matrix does not have an allocation on the device for the current range, create a new allocation and if specified, also copy sparse matrix data to device. Saves newly allocated ranges to m_deviceMemIndexPointers_CU so sparse matrix can keep track of where and what it has stored on devices.

Parameters

start	Pointer to first index element in range to be updated with device.
elems	Number of (non-zero) index elements.
deviceID	Integer specififying the device that should be synched with.
copy	Boolean value that tells whether to only allocate or also copy sparse matrix index data to device. True copies, False only allocates.

References skepu::DeviceMemPointer_CU< T >::copyHostToDevice(), and skepu::Environment< T >::getInstance().

Referenced by skepu::MapArray< MapArrayFunc >::CU().

Here is the call graph for this function: