The GpuView class is provides a view of some data allocated on the GPU Essenstially is only stores a pointer and a size. More...

#include <GpuView.hpp>

Classes
class	iterator
	Iterator class to make GpuViews more similar to std containers. More...

Public Types
using	value_type = T

Public Member Functions
	GpuView ()=default
	Default constructor that will initialize cublas and allocate 0 bytes of memory.

	GpuView (std::vector< T > &data)
	constructor based on std::vectors, this will make a view on the CPU More...

__host__ __device__ T &	operator[] (size_t idx)
	operator[] to retrieve a reference to an item in the buffer More...

__host__ __device__ T	operator[] (size_t idx) const
	operator[] to retrieve a copy of an item in the buffer More...

	GpuView (T *dataOnHost, size_t numberOfElements)
	GpuView allocates new GPU memory of size numberOfElements * sizeof(T) and copies numberOfElements from data. More...

	~GpuView ()=default
	~GpuView calls cudaFree

__host__ __device__ T *	data ()

__host__ __device__ const T *	data () const

__host__ __device__ bool	empty () const

__host__ __device__ T &	front ()

__host__ __device__ T &	back ()

__host__ __device__ T	front () const

__host__ __device__ T	back () const

void	copyFromHost (const T *dataPointer, size_t numberOfElements)
	copyFromHost copies numberOfElements from the CPU memory dataPointer More...

void	copyToHost (T *dataPointer, size_t numberOfElements) const
	copyFromHost copies numberOfElements to the CPU memory dataPointer More...

void	copyFromHost (const std::vector< T > &data)
	copyToHost copies data from an std::vector More...

void	copyToHost (std::vector< T > &data) const
	copyToHost copies data to an std::vector More...

__host__ __device__ size_t	size () const
	size returns the size (number of T elements) in the vector More...

std::vector< T >	asStdVector () const
	creates an std::vector of the same size and copies the GPU data to this std::vector More...

__host__ __device__ iterator	begin ()
	Get an iterator pointing to the first element of the buffer. More...

__host__ __device__ iterator	begin () const
	Get a const iterator pointing to the first element of the buffer. More...

__host__ __device__ iterator	end ()
	Get an iterator pointing to the address after the last element of the buffer. More...

__host__ __device__ iterator	end () const
	Get a const iterator pointing to the address after the last element of the buffer. More...

Detailed Description

template<typename T>
class Opm::gpuistl::GpuView< T >

The GpuView class is provides a view of some data allocated on the GPU Essenstially is only stores a pointer and a size.

This class supports being used from inside a CUDA/HIP Kernel. Implementations are placed in this headerfile for functions that may be called inside a kernel to avoid expensive RDC (relocatable device code)

The view will typically provide a view into a GpuBuffer and be able to manipulate the data within it

Parameters

T	Type of the data we store, typically int/float/double w/o const specifier

Constructor & Destructor Documentation

◆ GpuView() [1/2]

template<class T >

Opm::gpuistl::GpuView< T >::GpuView ( std::vector< T > & data )

constructor based on std::vectors, this will make a view on the CPU

Parameters

data	std vector to pr

◆ GpuView() [2/2]

template<typename T>

Opm::gpuistl::GpuView< T >::GpuView	(	T *	dataOnHost,
		size_t	numberOfElements
	)

inline

GpuView allocates new GPU memory of size numberOfElements * sizeof(T) and copies numberOfElements from data.

Note: This assumes the data is on the CPU.

Parameters

numberOfElements	number of T elements to allocate
dataOnHost	data on host/CPU

Member Function Documentation

◆ asStdVector()

template<typename T >

std::vector< T > Opm::gpuistl::GpuView< T >::asStdVector ( ) const

creates an std::vector of the same size and copies the GPU data to this std::vector

Returns: an std::vector containing the elements copied from the GPU.

◆ back() [1/2]

template<typename T>

__host__ __device__ T& Opm::gpuistl::GpuView< T >::back ( )

inline

Returns: fetch the last element in a GpuView

◆ back() [2/2]

template<typename T>

__host__ __device__ T Opm::gpuistl::GpuView< T >::back ( ) const

inline

Returns: fetch the last element in a GpuView

◆ begin() [1/2]

template<typename T>

__host__ __device__ iterator Opm::gpuistl::GpuView< T >::begin ( )

inline

Get an iterator pointing to the first element of the buffer.

Parameters

iterator to traverse the buffer

◆ begin() [2/2]

template<typename T>

__host__ __device__ iterator Opm::gpuistl::GpuView< T >::begin ( ) const

inline

Get a const iterator pointing to the first element of the buffer.

Parameters

iterator to traverse the buffer

◆ copyFromHost() [1/2]

template<class T >

void Opm::gpuistl::GpuView< T >::copyFromHost	(	const T *	dataPointer,
		size_t	numberOfElements
	)

copyFromHost copies numberOfElements from the CPU memory dataPointer

Parameters

dataPointer	raw pointer to CPU memory
numberOfElements	number of elements to copy

Note: This does synchronous transfer.; assumes that this view has numberOfElements elements

◆ copyFromHost() [2/2]

template<class T >

void Opm::gpuistl::GpuView< T >::copyFromHost ( const std::vector< T > & data )

copyToHost copies data from an std::vector

Parameters

data	the vector to copy from

Note: This does synchronous transfer.; This assumes that the size of this view is equal to the size of the input vector.

◆ copyToHost() [1/2]

template<class T >

void Opm::gpuistl::GpuView< T >::copyToHost	(	T *	dataPointer,
		size_t	numberOfElements
	)		const

copyFromHost copies numberOfElements to the CPU memory dataPointer

Parameters

dataPointer	raw pointer to CPU memory
numberOfElements	number of elements to copy

Note: This does synchronous transfer.; assumes that this view has numberOfElements elements

◆ copyToHost() [2/2]

template<class T >

void Opm::gpuistl::GpuView< T >::copyToHost ( std::vector< T > & data ) const

copyToHost copies data to an std::vector

Parameters

data	the vector to copy to

Note: This does synchronous transfer.; This assumes that the size of this view is equal to the size of the input vector.

◆ data() [1/2]

template<typename T>

__host__ __device__ T* Opm::gpuistl::GpuView< T >::data ( )

inline

Returns: the raw pointer to the GPU data

◆ data() [2/2]

template<typename T>

__host__ __device__ const T* Opm::gpuistl::GpuView< T >::data ( ) const

inline

Returns: the raw pointer to the GPU data

◆ empty()

template<typename T>

__host__ __device__ bool Opm::gpuistl::GpuView< T >::empty ( ) const

inline

Returns: true if the view has no elements, false otherwise

◆ end() [1/2]

template<typename T>

__host__ __device__ iterator Opm::gpuistl::GpuView< T >::end ( )

inline

Get an iterator pointing to the address after the last element of the buffer.

Parameters

iterator pointing to the first value after the end of the buffer

◆ end() [2/2]

template<typename T>

__host__ __device__ iterator Opm::gpuistl::GpuView< T >::end ( ) const

inline

Get a const iterator pointing to the address after the last element of the buffer.

Parameters

iterator pointing to the first value after the end of the buffer

◆ front() [1/2]

template<typename T>

__host__ __device__ T& Opm::gpuistl::GpuView< T >::front ( )

inline

Returns: fetch the first element in a GpuView

◆ front() [2/2]

template<typename T>

__host__ __device__ T Opm::gpuistl::GpuView< T >::front ( ) const

inline

Returns: fetch the first element in a GpuView

◆ operator[]() [1/2]

template<typename T>

__host__ __device__ T& Opm::gpuistl::GpuView< T >::operator[] ( size_t idx )

inline

operator[] to retrieve a reference to an item in the buffer

Parameters

idx	The index of the element

◆ operator[]() [2/2]

template<typename T>

__host__ __device__ T Opm::gpuistl::GpuView< T >::operator[] ( size_t idx ) const

inline

operator[] to retrieve a copy of an item in the buffer

Parameters

idx	The index of the element

◆ size()

template<typename T>

__host__ __device__ size_t Opm::gpuistl::GpuView< T >::size ( ) const

inline

size returns the size (number of T elements) in the vector

Returns: number of elements

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

opm/simulators/linalg/gpuistl/GpuView.hpp
opm/simulators/linalg/gpuistl/GpuView.cpp

Classes

Public Types

Public Member Functions

Detailed Description

template<typename T> class Opm::gpuistl::GpuView< T >

Constructor & Destructor Documentation

◆ GpuView() [1/2]

◆ GpuView() [2/2]

Member Function Documentation

◆ asStdVector()

◆ back() [1/2]

◆ back() [2/2]

◆ begin() [1/2]

◆ begin() [2/2]

◆ copyFromHost() [1/2]

◆ copyFromHost() [2/2]

◆ copyToHost() [1/2]

◆ copyToHost() [2/2]

◆ data() [1/2]

◆ data() [2/2]

◆ empty()

◆ end() [1/2]

◆ end() [2/2]

◆ front() [1/2]

◆ front() [2/2]

◆ operator[]() [1/2]

◆ operator[]() [2/2]

◆ size()

template<typename T>
class Opm::gpuistl::GpuView< T >