stormm::stmath::HpcMatrix< T > Class Template Reference

A matrix object that can traverse CPU and GPU memory. More...

#include <matrix.h>

Public Member Functions
HybridLabel	getLabel () const
	Produce the label for this matrix (obtained from the label of its primary contents)
void	resize (size_t new_rows, size_t new_cols)
	Change the number of rows and columns. Elements outside the bounds of the new matrix will be discarded. If the new matrix is larger than the old one, all elements will be preserved and new elements will be initialized to zero.
void	reset ()
	Empty the matrix and set its size to 0 by 0.
T	operator() (size_t row_idx, size_t col_idx, HybridTargetLevel tier=HybridTargetLevel::HOST)
	Element-wise access via the function call () operator. This entails two bounds checks.
T	atHost (size_t row_idx, size_t col_idx)
	Element-wise access to host data without bounds checks, following on the Armadillo naming conventions.
T	atDevice (size_t row_idx, size_t col_idx)
	Element-wise access to data on the GPU without bounds checks, following on the Armadillo naming conventions.
Hybrid< T >	colptr (size_t col_idx, const char *tag_in=nullptr)
	Get a Hybrid POINTER-kind object that is set directly to the column of interest.
	HpcMatrix (const char *tag_in=nullptr, HybridFormat format_in=default_hpc_format)
	Constructors must set public-facing reference variables in addition to private member variables.
	HpcMatrix (size_t rows_in, size_t cols_in, const char *tag_in=nullptr, HybridFormat format_in=default_hpc_format, MatrixFillMode fill_method=MatrixFillMode::VALUE, T fill_value=static_cast< T >(0))
	HpcMatrix (size_t rows_in, size_t cols_in, MatrixFillMode fill_method, T fill_value=static_cast< T >(0), const char *tag_in=nullptr, HybridFormat format_in=default_hpc_format)
	HpcMatrix (size_t rows_in, size_t cols_in, MatrixFillMode fill_method, Ran2Generator *matrng, const char *tag_in=nullptr, HybridFormat format_in=default_hpc_format)
const T *	memptr (HybridTargetLevel tier=HybridTargetLevel::HOST) const
	Pointer to the memory itself, on the host or the device. This takes its name from the Armadillo vec, mat, and cube classes.
T *	memptr (HybridTargetLevel tier=HybridTargetLevel::HOST)

Public Attributes
const size_t &	n_rows
const size_t &	n_cols
const size_t &	n_elem

Detailed Description

template<typename T>
class stormm::stmath::HpcMatrix< T >

A matrix object that can traverse CPU and GPU memory.

Constructor & Destructor Documentation

HpcMatrix()

template<typename T>

stormm::stmath::HpcMatrix< T >::HpcMatrix	(	const char *	tag_in = nullptr,
		HybridFormat	format_in = default_hpc_format )

Constructors must set public-facing reference variables in addition to private member variables.

Overloaded:

• Initialize an empty matrix with zero rows and zero columns
• Initialize a matrix with a given numer of rows and columns, all elements set to zero

Parameters

rows_in	The number of rows to allocate
cols_in	The number of columns to allocate
fill_method	Manner in which to fill the indices
fill_value	A scalar value for initializing all elements if fill_method is VALUE, or the random seed for initializing a random number generator if fill_method is RANDU or RANDN

Member Function Documentation

atDevice()

template<typename T>

T stormm::stmath::HpcMatrix< T >::atDevice	(	size_t	row_idx,
		size_t	col_idx )

Element-wise access to data on the GPU without bounds checks, following on the Armadillo naming conventions.

Parameters

row_idx	Index of the row to access, starting at 0
col_idx	Index of the column to access, starting at 0

atHost()

template<typename T>

T stormm::stmath::HpcMatrix< T >::atHost	(	size_t	row_idx,
		size_t	col_idx )

Element-wise access to host data without bounds checks, following on the Armadillo naming conventions.

Parameters

row_idx	Index of the row to access, starting at 0
col_idx	Index of the column to access, starting at 0

colptr()

template<typename T>

Hybrid< T > stormm::stmath::HpcMatrix< T >::colptr	(	size_t	col_idx,
		const char *	tag_in = nullptr )

Get a Hybrid POINTER-kind object that is set directly to the column of interest.

Parameters

col_idx	The column of interest
tag_in	A note to help identify the column pointer

memptr()

template<typename T>

const T * stormm::stmath::HpcMatrix< T >::memptr

(

HybridTargetLevel

tier = HybridTargetLevel::HOST

)

const

Pointer to the memory itself, on the host or the device. This takes its name from the Armadillo vec, mat, and cube classes.

Overloaded:

• Get a const pointer if the object is const
• Get a pointer that can modify the underlying data if the object is non-const

Parameters

tier	Indicator of whether to offer a pointer to host or GPU device memory

operator()()

template<typename T>

T stormm::stmath::HpcMatrix< T >::operator()	(	size_t	row_idx,
		size_t	col_idx,
		HybridTargetLevel	tier = HybridTargetLevel::HOST )

Element-wise access via the function call () operator. This entails two bounds checks.

Parameters

row_idx	Index of the row to access, starting at 0
col_idx	Index of the column to access, starting at 0
tier	Indicator of whether to access data on the host or on the GPU

resize()

template<typename T>

void stormm::stmath::HpcMatrix< T >::resize	(	size_t	new_rows,
		size_t	new_cols )

Change the number of rows and columns. Elements outside the bounds of the new matrix will be discarded. If the new matrix is larger than the old one, all elements will be preserved and new elements will be initialized to zero.

Parameters

new_rows	New number of rows to allocate
new_cols	New number of columns to allocate

---

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

• src/Math/matrix.h