CUDA-Device-global memory on a single device (not accessible from the host) More...

Typedefs
using	address_t = CUdeviceptr
	The numeric type which can represent the range of memory addresses on a CUDA device. More...

using	unique_region = memory::unique_region< detail_::deleter >
	A unique region of device-global memory.

Functions
void	free (void *ptr)
	Free a region of device-side memory (regardless of how it was allocated)

void	free (region_t region)
	Free a region of device-side memory (regardless of how it was allocated) More...

region_t	allocate (const context_t &context, size_t size_in_bytes)
	Allocate device-side memory on a CUDA device context. More...

region_t	allocate (const device_t &device, size_t size_in_bytes)
	Allocate device-side memory on a CUDA device. More...

template<typename T >
void	typed_set (T *start, const T &value, size_t num_elements, optional_ref< const stream_t > stream={})
	Sets consecutive elements of a region of memory to a fixed value of some width. More...

void	set (void *start, int byte_value, size_t num_bytes, optional_ref< const stream_t > stream={})
	Sets all bytes in a region of memory to a fixed value. More...

void	set (region_t region, int byte_value, optional_ref< const stream_t > stream={})
	Sets all bytes in a region of memory to a fixed value. More...

void	zero (void *start, size_t num_bytes, optional_ref< const stream_t > stream={})
	Sets all bytes in a region of memory to 0 (zero) More...

void	zero (region_t region, optional_ref< const stream_t > stream={})
	Sets all bytes in a region of memory to 0 (zero) More...

template<typename T >
void	zero (T *ptr, optional_ref< const stream_t > stream={})
	Sets all bytes of a single pointed-to value to 0. More...

template<typename T >
unique_span< T >	make_unique_span (const context_t &context, size_t size)
	Allocate memory for a consecutive sequence of typed elements in device-global memory. More...

template<typename T >
unique_span< T >	make_unique_span (const device_t &device, size_t size)
	Allocate memory for a consecutive sequence of typed elements in device-global memory. More...

template<typename T >
unique_span< T >	make_unique_span (size_t size)
	Allocate memory for a consecutive sequence of typed elements in device-global memory. More...

unique_region	make_unique_region (const context_t &context, size_t num_bytes)
	Allocate a region in device-global memory. More...

unique_region	make_unique_region (const device_t &device, size_t num_elements)
	Create a variant of ::std::unique_pointer for an array in device-global memory. More...

unique_region	make_unique_region (size_t num_elements)
	Create a variant of ::std::unique_pointer for an array in device-global memory on the current device. More...

address_t	address (const void *device_ptr) noexcept

address_t	address (memory::const_region_t region) noexcept

Detailed Description

CUDA-Device-global memory on a single device (not accessible from the host)

Typedef Documentation

◆ address_t

using cuda::memory::device::address_t = typedef CUdeviceptr

The numeric type which can represent the range of memory addresses on a CUDA device.

As these addresses are typically just part of the single, unified all-system memory space, this should be the same type as a system memory address' numeric equivalent.

Function Documentation

◆ address() [1/2]

address_t cuda::memory::device::address ( const void * device_ptr )

inlinenoexcept

Returns: a cast of a proper pointer into a numeric address in device memory space (which is usually just a part of the unified all-system memory space)

Note: Typically, this is just a reinterpretation of the same value.

◆ address() [2/2]

address_t cuda::memory::device::address ( memory::const_region_t region )

inlinenoexcept

Returns: The numeric address of the beginning of a memory region

Note: Typically, this is just a reinterpretation of the same value.

◆ allocate() [1/2]

region_t cuda::memory::device::allocate	(	const context_t &	context,
		size_t	size_in_bytes
	)

inline

Allocate device-side memory on a CUDA device context.

Note: The CUDA memory allocator guarantees alignment "suitabl[e] for any kind of variable" (CUDA 9.0 Runtime API documentation), and the CUDA programming guide guarantees since at least version 5.0 that the minimum allocation is 256 bytes.

Exceptions

cuda::runtime_error if allocation fails for any reason

Parameters

context	the context in which to allocate memory
size_in_bytes	the amount of global device memory to allocate

Returns: a pointer to the allocated stretch of memory (only usable within context)

◆ allocate() [2/2]

region_t cuda::memory::device::allocate	(	const device_t &	device,
		size_t	size_in_bytes
	)

inline

Allocate device-side memory on a CUDA device.

Note: The CUDA memory allocator guarantees alignment "suitabl[e] for any kind of variable" (CUDA 9.0 Runtime API documentation), and the CUDA programming guide guarantees since at least version 5.0 that the minimum allocation is 256 bytes.

Exceptions

cuda::runtime_error if allocation fails for any reason

Parameters

device	the device on which to allocate memory
size_in_bytes	the amount of global device memory to allocate

Returns: a pointer to the allocated stretch of memory (only usable on device)

◆ free()

void cuda::memory::device::free ( region_t region )

inline

Free a region of device-side memory (regardless of how it was allocated)

◆ make_unique_region() [1/3]

unique_region cuda::memory::device::make_unique_region	(	const context_t &	context,
		size_t	num_bytes
	)

inline

Allocate a region in device-global memory.

Parameters

context	The context within which (and in the device global memory of which) to make the allocation
num_bytes	Size of the region to be allocated, in bytes

Returns: An owning RAII/CADRe object for the allocated memory region

◆ make_unique_region() [2/3]

unique_region cuda::memory::device::make_unique_region	(	const device_t &	device,
		size_t	num_elements
	)

inline

Create a variant of ::std::unique_pointer for an array in device-global memory.

Allocate a region in device-global memory.

Template Parameters

T	an array type; not the type of individual elements

Parameters

device	on which to construct the array of elements
num_elements	the number of elements to allocate

Returns: an ::std::unique_ptr pointing to the constructed T array

Parameters

device The device in the global memory of which to make the allocation

Returns: An owning RAII/CADRe object for the allocated memory region

◆ make_unique_region() [3/3]

unique_region cuda::memory::device::make_unique_region ( size_t num_elements )

inline

Create a variant of ::std::unique_pointer for an array in device-global memory on the current device.

Allocate a region of managed memory, accessible both from CUDA devices and from the CPU.

Allocate a region in device-global memory within the primary context of the current CUDA device.

Note: The allocation will be made in the device's primary context - which will be created if it has not yet been.

Template Parameters

T	an array type; not the type of individual elements

Parameters

num_elements the number of elements to allocate

Returns: an ::std::unique_ptr pointing to the constructed T array

Parameters

device The device in the global memory of which to make the allocation

Returns: An owning RAII/CADRe object for the allocated memory region

◆ make_unique_span() [1/3]

template<typename T >

unique_span< T > cuda::memory::device::make_unique_span	(	const context_t &	context,
		size_t	size
	)

Allocate memory for a consecutive sequence of typed elements in device-global memory.

Template Parameters

T	type of the individual elements in the allocated sequence

Parameters

context	The CUDA device context in which to make the allocation.
size	the number of elements to allocate

Returns: A unique_span which owns the allocated memory (and will release said

Note: This function is somewhat similar to ::std:: make_unique_for_overwrite(), except that the returned value is not "just" a unique pointer, but also has a size. It is also similar to {cuda::device::make_unique_region}, except that the allocation is conceived as typed elements.; Typically, this is used for trivially-constructible elements, for which reason the non-construction of individual elements should not pose a problem. But - let the user beware.

◆ make_unique_span() [2/3]

template<typename T >

unique_span< T > cuda::memory::device::make_unique_span	(	const device_t &	device,
		size_t	num_elements
	)

Allocate memory for a consecutive sequence of typed elements in device-global memory.

Allocate (but do) device-global memory.

Template Parameters

T	type of the individual elements in the allocated sequence

Parameters

context	The CUDA device context in which to make the allocation.
size	the number of elements to allocate

Returns: A unique_span which owns the allocated memory (and will release said

Note: This function is somewhat similar to ::std:: make_unique_for_overwrite(), except that the returned value is not "just" a unique pointer, but also has a size. It is also similar to {cuda::device::make_unique_region}, except that the allocation is conceived as typed elements.; Typically, this is used for trivially-constructible elements, for which reason the non-construction of individual elements should not pose a problem. But - let the user beware.

Parameters

device The CUDA device in whose primary context to make the allocation.

Template Parameters

T	an array type; not the type of individual elements

Parameters

device	on which to construct the array of elements
num_elements	the number of elements to allocate

Returns: an ::std::unique_ptr pointing to the constructed T array

◆ make_unique_span() [3/3]

template<typename T >

unique_span< T > cuda::memory::device::make_unique_span ( size_t num_elements )

Allocate memory for a consecutive sequence of typed elements in device-global memory.

Create a variant of ::std::unique_pointer for an array in device-global memory on the current device.

Template Parameters

T	type of the individual elements in the allocated sequence

Parameters

context	The CUDA device context in which to make the allocation.
size	the number of elements to allocate

Returns: A unique_span which owns the allocated memory (and will release said

Note: This function is somewhat similar to ::std:: make_unique_for_overwrite(), except that the returned value is not "just" a unique pointer, but also has a size. It is also similar to {cuda::device::make_unique_region}, except that the allocation is conceived as typed elements.; Typically, this is used for trivially-constructible elements, for which reason the non-construction of individual elements should not pose a problem. But - let the user beware.; The current device's primary context will be used (not the current context).; The allocation will be made in the device's primary context - which will be created if it has not yet been.

Template Parameters

T	an array type; not the type of individual elements

Parameters

num_elements the number of elements to allocate

Returns: an ::std::unique_ptr pointing to the constructed T array

◆ set() [1/2]

void cuda::memory::device::set	(	void *	start,
		int	byte_value,
		size_t	num_bytes,
		optional_ref< const stream_t >	stream = `{}`
	)

inline

Sets all bytes in a region of memory to a fixed value.

Note: The equivalent of ::std::memset for CUDA device-side memory

Parameters

byte_value	value to set the memory region to
start	starting address of the memory region to set, in a CUDA device's global memory
num_bytes	size of the memory region in bytes
stream	an stream on which to schedule the operation; may be omitted

Asynchronously sets all bytes in a stretch of memory to a single value

Note: asynchronous version of memory::set(void*, int, size_t)

Parameters

start	starting address of the memory region to set, in a CUDA device's global memory
byte_value	value to set the memory region to
num_bytes	size of the memory region in bytes
stream	A stream on which to schedule this action; may be omitted.

◆ set() [2/2]

void cuda::memory::device::set	(	region_t	region,
		int	byte_value,
		optional_ref< const stream_t >	stream = `{}`
	)

inline

Sets all bytes in a region of memory to a fixed value.

Note: The equivalent of ::std::memset for CUDA device-side memory

Parameters

byte_value	value to set the memory region to
region	a region to zero-out, in a CUDA device's global memory
stream	A stream on which to schedule this action; may be omitted.

◆ typed_set()

template<typename T >

void cuda::memory::device::typed_set	(	T *	start,
		const T &	value,
		size_t	num_elements,
		optional_ref< const stream_t >	stream = `{}`
	)

inline

Sets consecutive elements of a region of memory to a fixed value of some width.

Note: A generalization of set(), for different-size units.

Template Parameters

T	An unsigned integer type of size 1, 2, 4 or 8

Parameters

start	The first location to set to `value` ; must be properly aligned.
value	A (properly aligned) value to set T-elements to.
num_elements	The number of type-T elements (i.e. not necessarily the number of bytes).
stream	A stream on which to schedule this action; may be omitted.

◆ zero() [1/3]

void cuda::memory::device::zero	(	void *	start,
		size_t	num_bytes,
		optional_ref< const stream_t >	stream = `{}`
	)

inline

Sets all bytes in a region of memory to 0 (zero)

Asynchronously sets all bytes in a stretch of memory to 0.

Parameters

start	the beginning of a region of memory to zero-out, accessible within a CUDA device's global memory
num_bytes	the size in bytes of the region of memory to zero-out
stream	A stream on which to schedule this action; may be omitted.
start	starting address of the memory region to set, in a CUDA device's global memory
num_bytes	size of the memory region in bytes
stream	stream on which to schedule this action
stream	A stream on which to enqueue the operation; may be omitted.

◆ zero() [2/3]

void cuda::memory::device::zero	(	region_t	region,
		optional_ref< const stream_t >	stream = `{}`
	)

inline

Sets all bytes in a region of memory to 0 (zero)

Parameters

region	the memory region to zero-out, accessible as a part of a CUDA device's global memory
stream	A stream on which to schedule this action; may be omitted.

◆ zero() [3/3]

template<typename T >

void cuda::memory::device::zero	(	T *	ptr,
		optional_ref< const stream_t >	stream = `{}`
	)

inline

Sets all bytes of a single pointed-to value to 0.

Parameters

ptr	pointer to a value of a certain type, accessible within in a CUDA device's global memory
stream	an existing stream on which to schedule this action; may be omitted

Typedefs

Functions

Detailed Description

Typedef Documentation

◆ address_t

Function Documentation

◆ address() [1/2]

◆ address() [2/2]

◆ allocate() [1/2]

◆ allocate() [2/2]

◆ free()

◆ make_unique_region() [1/3]

◆ make_unique_region() [2/3]

◆ make_unique_region() [3/3]

◆ make_unique_span() [1/3]

◆ make_unique_span() [2/3]

◆ make_unique_span() [3/3]

◆ set() [1/2]

◆ set() [2/2]

◆ typed_set()

◆ zero() [1/3]

◆ zero() [2/3]

◆ zero() [3/3]