Utilities

Files
file	aligned.hpp
file	prefetch.hpp

Functions
bool	rmm::is_pow2 (std::size_t value) noexcept
	Returns whether or not value is a power of 2. More...
bool	rmm::is_supported_alignment (std::size_t alignment) noexcept
	Returns whether or not alignment is a valid memory alignment. More...
std::size_t	rmm::align_up (std::size_t value, std::size_t alignment) noexcept
	Align up to nearest multiple of specified power of 2. More...
std::size_t	rmm::align_down (std::size_t value, std::size_t alignment) noexcept
	Align down to the nearest multiple of specified power of 2. More...
bool	rmm::is_aligned (std::size_t value, std::size_t alignment) noexcept
	Checks whether a value is aligned to a multiple of a specified power of 2. More...
bool	rmm::is_pointer_aligned (void *ptr, std::size_t alignment=CUDA_ALLOCATION_ALIGNMENT) noexcept
	Checks whether the provided pointer is aligned to a specified alignment. More...
void	rmm::prefetch (void const *ptr, std::size_t size, rmm::cuda_device_id device, rmm::cuda_stream_view stream)
	Prefetch memory to the specified device on the specified stream. More...
template<typename T >
void	rmm::prefetch (cuda::std::span< T const > data, rmm::cuda_device_id device, rmm::cuda_stream_view stream)
	Prefetch a span of memory to the specified device on the specified stream. More...

Variables
static constexpr std::size_t	rmm::RMM_DEFAULT_HOST_ALIGNMENT {alignof(std::max_align_t)}
	Default alignment used for host memory allocated by RMM. More...
static constexpr std::size_t	rmm::CUDA_ALLOCATION_ALIGNMENT {256}
	Default alignment used for CUDA memory allocation.

Detailed Description

Function Documentation

align_down()

std::size_t rmm::align_down ( std::size_t  value, std::size_t  alignment  )

noexcept

Align down to the nearest multiple of specified power of 2.

Parameters

[in]	value	value to align
[in]	alignment	amount, in bytes, must be a power of 2

Returns

the aligned value

align_up()

std::size_t rmm::align_up ( std::size_t  value, std::size_t  alignment  )

noexcept

Align up to nearest multiple of specified power of 2.

Parameters

[in]	value	value to align
[in]	alignment	amount, in bytes, must be a power of 2

Returns

the aligned value

is_aligned()

bool rmm::is_aligned ( std::size_t  value, std::size_t  alignment  )

noexcept

Checks whether a value is aligned to a multiple of a specified power of 2.

Parameters

[in]	value	value to check for alignment
[in]	alignment	amount, in bytes, must be a power of 2

Returns

true if aligned

is_pointer_aligned()

bool rmm::is_pointer_aligned ( void *  ptr, std::size_t  alignment = CUDA_ALLOCATION_ALIGNMENT  )

noexcept

Checks whether the provided pointer is aligned to a specified alignment.

Parameters

[in]	ptr	pointer to check for alignment
[in]	alignment	required alignment in bytes, must be a power of 2

Returns

true if the pointer is aligned

is_pow2()

bool rmm::is_pow2 ( std::size_t  value )

noexcept

Returns whether or not value is a power of 2.

Parameters

[in]

value

value to check.

Returns

True if the input is a power of two with non-negative integer exponent, false otherwise.

is_supported_alignment()

bool rmm::is_supported_alignment ( std::size_t  alignment )

noexcept

Returns whether or not alignment is a valid memory alignment.

Parameters

[in]

alignment

to check

Returns

True if the alignment is valid, false otherwise.

prefetch() [1/2]

template<typename T >

void rmm::prefetch	(	cuda::std::span< T const >	data,
		rmm::cuda_device_id	device,
		rmm::cuda_stream_view	stream
	)

Prefetch a span of memory to the specified device on the specified stream.

This function is a no-op if the buffer is not backed by CUDA managed memory.

Exceptions

rmm::cuda_error

if the prefetch fails.

Parameters

data	The span to prefetch
device	The device to prefetch to
stream	The stream to use for the prefetch

prefetch() [2/2]

void rmm::prefetch	(	void const *	ptr,
		std::size_t	size,
		rmm::cuda_device_id	device,
		rmm::cuda_stream_view	stream
	)

Prefetch memory to the specified device on the specified stream.

This function is a no-op if the pointer is not to CUDA managed memory or if concurrent managed access is not supported.

Exceptions

rmm::cuda_error

if the prefetch fails.

Parameters

ptr	The pointer to the memory to prefetch
size	The number of bytes to prefetch
device	The device to prefetch to
stream	The stream to use for the prefetch

Variable Documentation

RMM_DEFAULT_HOST_ALIGNMENT

constexpr std::size_t rmm::RMM_DEFAULT_HOST_ALIGNMENT {alignof(std::max_align_t)}

staticconstexpr

Default alignment used for host memory allocated by RMM.

Deprecated:

This constant is deprecated in 25.12 and will be removed in 26.02.