Detailed Description

Generic versions (w.r.t. the different address spaces) of memory managing functions as well as other tools related to memory management.

Concepts
concept	nda::mem::Allocator
	Check if a given type satisfies the allocator concept.

concept	nda::mem::Handle
	Check if a given type satisfies the memory handle concept.

concept	nda::mem::OwningHandle
	Check if a given type satisfies the owning memory handle concept.

Classes
struct	nda::mem::aligner< T, Al >
	Wraps an arbitrary type to have a specified alignment. More...

struct	nda::mem::do_not_initialize_t
	Tag used in constructors to indicate that the memory should not be initialized. More...

struct	nda::mem::init_zero_t
	Tag used in constructors to indicate that the memory should be initialized to zero. More...

Macros
#define	device_error_check(ARG1, ARG2)
	Trigger a compilation error every time the nda::device_error_check function is called.

Functions
template<bool flag = false>
void	nda::compile_error_no_gpu ()
	Trigger a compilation error in case GPU specific functionality is used without configuring the project with GPU support.

template<AddressSpace AdrSp>
void	nda::mem::free (void *p)
	Call the correct `free` function based on the given address space.

template<AddressSpace AdrSp>
void *	nda::mem::malloc (size_t size)
	Call the correct `malloc` function based on the given address space.

template<AddressSpace DestAdrSp, AddressSpace SrcAdrSp>
void	nda::mem::memcpy (void dest, void const src, size_t count)
	Call the correct `memcpy` function based on the given address spaces.

template<AddressSpace DestAdrSp, AddressSpace SrcAdrSp>
void	nda::mem::memcpy2D (void dest, size_t dpitch, const void src, size_t spitch, size_t width, size_t height)
	Call CUDA's `cudaMemcpy2D` function or simulate its behavior on the `Host` based on the given address spaces.

template<AddressSpace AdrSp>
void	nda::mem::memset (void *p, int value, size_t count)
	Call the correct `memset` function based on the given address space.

template<AddressSpace AdrSp>
void	nda::mem::memset2D (void *ptr, size_t pitch, int value, size_t width, size_t height)
	Call CUDA's `cudaMemset2D` function or simulate its behavior on the `Host` based on the given address space.

Variables
static constexpr do_not_initialize_t	nda::mem::do_not_initialize {}
	Instance of nda::mem::do_not_initialize_t.

static constexpr bool	nda::have_cuda = false
	Constexpr variable that is true if the project is configured with CUDA support.

static constexpr bool	nda::have_device = false
	Constexpr variable that is true if the project is configured with GPU support.

static constexpr bool	nda::mem::init_dcmplx = true
	Should we initialize memory for complex double types to zero.

static constexpr init_zero_t	nda::mem::init_zero {}
	Instance of nda::mem::init_zero_t.

Function Documentation

◆ free()

template<AddressSpace AdrSp>

void nda::mem::free ( void * p )

#include <nda/mem/malloc.hpp>

Call the correct free function based on the given address space.

It makes the following function calls depending on the address space:

std::free for Host.
cudaFree for Device and Unified.

Template Parameters

AdrSp nda::mem::AddressSpace.

Parameters

p	Pointer to the memory to be freed.

Definition at line 75 of file malloc.hpp.

◆ malloc()

template<AddressSpace AdrSp>

void * nda::mem::malloc ( size_t size )

#include <nda/mem/malloc.hpp>

Call the correct malloc function based on the given address space.

It makes the following function calls depending on the address space:

std::malloc for Host.
cudaMalloc for Device.
cudaMallocManaged for Unified.

Template Parameters

AdrSp nda::mem::AddressSpace.

Parameters

size	Size in bytes to be allocated.

Returns: Pointer to the allocated memory.

Definition at line 49 of file malloc.hpp.

◆ memcpy()

template<AddressSpace DestAdrSp, AddressSpace SrcAdrSp>

void nda::mem::memcpy	(	void *	dest,
		void const *	src,
		size_t	count )

#include <nda/mem/memcpy.hpp>

Call the correct memcpy function based on the given address spaces.

It makes the following function calls depending on the address spaces:

std::memcpy if both address spaces are Host.
cudaMemcpy for all other combinations.

Template Parameters

DestAdrSp	nda::mem::AddressSpace of the destination.
SrcAdrSp	nda::mem::AddressSpace of the source.

Parameters

dest	Pointer to the destination memory.
src	Pointer to the source memory.
count	Size in bytes to copy.

Definition at line 51 of file memcpy.hpp.

◆ memcpy2D()

template<AddressSpace DestAdrSp, AddressSpace SrcAdrSp>

void nda::mem::memcpy2D	(	void *	dest,
		size_t	dpitch,
		const void *	src,
		size_t	spitch,
		size_t	width,
		size_t	height )

#include <nda/mem/memcpy.hpp>

Call CUDA's cudaMemcpy2D function or simulate its behavior on the Host based on the given address spaces.

Copies a matrix (height rows of width bytes each) from the memory area pointed to by src to the memory area pointed to by dest. dpitch and spitch are the widths in memory in bytes of the 2D arrays pointed to by dest and src, including any padding added to the end of each row. The memory areas may not overlap. width must not exceed either dpitch or spitch.

If both address spaces are Host, it simulates the behavior of CUDA's cudaMemcpy2D function by making multiple calls to std::memcpy.

Template Parameters

DestAdrSp	nda::mem::AddressSpace of the destination.
SrcAdrSp	nda::mem::AddressSpace of the source.

Parameters

dest	Pointer to the destination memory.
dpitch	Pitch of destination memory
src	Pointer to the source memory.
spitch	Pitch of source memory.
width	Width of matrix transfer (columns in bytes).
height	Height of matrix transfer (rows).

Definition at line 84 of file memcpy.hpp.

◆ memset()

template<AddressSpace AdrSp>

void nda::mem::memset	(	void *	p,
		int	value,
		size_t	count )

#include <nda/mem/memset.hpp>

Call the correct memset function based on the given address space.

It makes the following function calls depending on the address spaces:

std::memset for Host.
cudaMemset for Device and Unified.

Template Parameters

AdrSp nda::mem::AddressSpace.

Parameters

p	Pointer to the memory to be set.
value	Value to set for each byte of specified memory.
count	Size in bytes to be set.

Definition at line 49 of file memset.hpp.

◆ memset2D()

template<AddressSpace AdrSp>

void nda::mem::memset2D	(	void *	ptr,
		size_t	pitch,
		int	value,
		size_t	width,
		size_t	height )

#include <nda/mem/memset.hpp>

Call CUDA's cudaMemset2D function or simulate its behavior on the Host based on the given address space.

Sets each byte of a matrix (height rows of width bytes each) pointed to by dest to a specified value. pitch is the width in memory in bytes of the 2D array pointed to by dest, including any padding added to the end of each row.

If the address space is Host, it simulates the behavior of CUDA's cudaMemset2D function by making multiple calls to std::memset.

Template Parameters

AdrSp nda::mem::AddressSpace.

Parameters

ptr	Pointer to the memory to be set.
pitch	Pitch in bytes of the memory (unused if height is 1).
value	Value to set for each byte of specified memory
width	Width of matrix to set (columns in bytes).
height	Height of matrix to set (rows).

Definition at line 78 of file memset.hpp.

Detailed Description

Concepts

Classes

Macros

Functions

Variables

Function Documentation

◆ free()

◆ malloc()

◆ memcpy()

◆ memcpy2D()

◆ memset()

◆ memset2D()