pfc::gpu::GPUVector< T > Class Template Reference

Simple GPU memory container (RAII) More...

#include <gpu_vector.hpp>

Public Member Functions
	GPUVector (size_t size)
	Construct a GPUVector with the specified size.
	~GPUVector ()
	Destructor - automatically frees GPU memory.
	GPUVector (const GPUVector &)=delete
GPUVector &	operator= (const GPUVector &)=delete
	GPUVector (GPUVector &&other) noexcept
	Move constructor.
GPUVector &	operator= (GPUVector &&other) noexcept
	Move assignment operator.
T *	data ()
	Get pointer to GPU memory.
const T *	data () const
	Get const pointer to GPU memory.
size_t	size () const
	Get the number of elements.
bool	empty () const
	Check if the vector is empty.
void	copy_from_host (const std::vector< T > &host_data)
	Copy data from host (CPU) to device (GPU)
void	copy_to_host (std::vector< T > &host_data) const
	Copy data from device (GPU) to host (CPU)
std::vector< T >	to_host () const
	Copy data from device to host and return as vector.

Detailed Description

template<typename T>
class pfc::gpu::GPUVector< T >

Simple GPU memory container (RAII)

Similar to std::vector but memory lives on GPU. Fixed size (no resizing).

Template Parameters

T	Element type (must be trivially copyable)

Note

This class only works when CUDA is enabled at compile time. On systems without CUDA, the class exists but operations will fail at runtime (or compile-time if properly guarded).

Warning

Do not use this class directly on systems without CUDA. Always check OpenPFC_ENABLE_CUDA before using.

Constructor & Destructor Documentation

GPUVector()

template<typename T >

pfc::gpu::GPUVector< T >::GPUVector ( size_t  size )

inlineexplicit

Construct a GPUVector with the specified size.

Allocates GPU memory for size elements of type T.

Parameters

size	Number of elements to allocate

Exceptions

std::runtime_error

if GPU memory allocation fails

Note

On systems without CUDA, this will throw at runtime. Always guard usage with ifdef OpenPFC_ENABLE_CUDA.

Here is the call graph for this function:

Member Function Documentation

copy_from_host()

template<typename T >

void pfc::gpu::GPUVector< T >::copy_from_host ( const std::vector< T > &  host_data )

inline

Copy data from host (CPU) to device (GPU)

Parameters

host_data

Source data on CPU

Exceptions

std::runtime_error

if sizes don't match or copy fails

copy_to_host()

template<typename T >

void pfc::gpu::GPUVector< T >::copy_to_host ( std::vector< T > &  host_data ) const

inline

Copy data from device (GPU) to host (CPU)

Parameters

host_data

Destination vector (will be resized if needed)

Here is the caller graph for this function:

data()

template<typename T >

T * pfc::gpu::GPUVector< T >::data ( )

inline

Get pointer to GPU memory.

Returns

Pointer to device memory (nullptr if size is 0)

Warning

This pointer is only valid on the GPU. Do not dereference on the CPU. Use copy_from_host/copy_to_host for CPU access.

to_host()

template<typename T >

std::vector< T > pfc::gpu::GPUVector< T >::to_host ( ) const

inline

Copy data from device to host and return as vector.

Returns

std::vector<T> containing the data from GPU

Here is the call graph for this function:

---

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

• openpfc/gpu/gpu_vector.hpp