heterogeneous accelerator memory resource More...

Classes
class	activate_cuda_device

class	activate_device

class	activate_hip_device

class	activate_openmp_device

struct	array_interface_tt
	traits for Numpy's array interface protocol More...

struct	array_interface_tt< char >

struct	array_interface_tt< double >

struct	array_interface_tt< float >

struct	array_interface_tt< int >

struct	array_interface_tt< long >

struct	array_interface_tt< long long >

struct	array_interface_tt< short >

struct	array_interface_tt< unsigned char >

struct	array_interface_tt< unsigned int >

struct	array_interface_tt< unsigned long >

struct	array_interface_tt< unsigned long long >

struct	array_interface_tt< unsigned short >

class	buffer
	A technology agnostic buffer that manages memory on CPUs, GPUs, and accelerators. More...

class	buffer_handle

struct	buffer_handle_tt
	type traits for constructing SWIG wrapped objects More...

struct	buffer_handle_tt< char >

struct	buffer_handle_tt< double >

struct	buffer_handle_tt< float >

struct	buffer_handle_tt< int >

struct	buffer_handle_tt< long >

struct	buffer_handle_tt< long long >

struct	buffer_handle_tt< short >

struct	buffer_handle_tt< unsigned char >

struct	buffer_handle_tt< unsigned int >

struct	buffer_handle_tt< unsigned long >

struct	buffer_handle_tt< unsigned long long >

struct	buffer_handle_tt< unsigned short >

struct	cuda_malloc_allocator

struct	cuda_malloc_allocator< T, std::enable_if< std::is_arithmetic< T >::value >::type >

struct	cuda_malloc_allocator< T, std::enable_if<!std::is_arithmetic< T >::value >::type >

struct	cuda_malloc_async_allocator

struct	cuda_malloc_async_allocator< T, std::enable_if< std::is_arithmetic< T >::value >::type >

struct	cuda_malloc_async_allocator< T, std::enable_if<!std::is_arithmetic< T >::value >::type >

class	cuda_malloc_async_deleter
	a deleter for arrays allocated with the cuda_malloc_async_allocator More...

class	cuda_malloc_async_deleter< T, std::enable_if< std::is_arithmetic< T >::value >::type >
	A deleter for arrays allocated with the cuda_malloc_async_allocator, specialized for numbers. More...

class	cuda_malloc_async_deleter< T, std::enable_if<!std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with the cuda_malloc_async_allocator, specialized for objects More...

class	cuda_malloc_deleter
	a deleter for arrays allocated with cudaMalloc More...

class	cuda_malloc_deleter< T, std::enable_if< std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with cudaMalloc, specialized for numbers More...

class	cuda_malloc_deleter< T, std::enable_if<!std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with cudaMalloc, specialized for objects More...

struct	cuda_malloc_host_allocator

struct	cuda_malloc_host_allocator< T, std::enable_if< std::is_arithmetic< T >::value >::type >

struct	cuda_malloc_host_allocator< T, std::enable_if<!std::is_arithmetic< T >::value >::type >

class	cuda_malloc_host_deleter
	a deleter for arrays allocated with cudaMallocHost More...

class	cuda_malloc_host_deleter< T, std::enable_if< std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with cudaMallocHost, specialized for numbers More...

class	cuda_malloc_host_deleter< T, std::enable_if<!std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with cudaMallocHost, specialized for objects More...

struct	cuda_malloc_uva_allocator
	a class for allocating arrays with cuda_malloc_uva More...

struct	cuda_malloc_uva_allocator< T, std::enable_if< std::is_arithmetic< T >::value >::type >
	a class for allocating arrays with cuda_malloc_uva, specialized for numbers More...

struct	cuda_malloc_uva_allocator< T, std::enable_if<!std::is_arithmetic< T >::value >::type >
	a class for allocating arrays with cuda_malloc_uva, specialized for objects More...

class	cuda_malloc_uva_deleter
	a deleter for arrays allocated with cuda_malloc_uva More...

class	cuda_malloc_uva_deleter< T, typename std::enable_if< std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with cuda_malloc_uva, specialized for numbers More...

class	cuda_malloc_uva_deleter< T, typename std::enable_if<!std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with cuda_malloc_uva, specialized for objects More...

class	gil_state
	A RAII helper for managing the Python GIL. More...

struct	hip_malloc_allocator
	a class for allocating arrays with hip_malloc More...

struct	hip_malloc_allocator< T, std::enable_if< std::is_arithmetic< T >::value >::type >
	a class for allocating arrays with hip_malloc, specialized for numbers More...

struct	hip_malloc_allocator< T, std::enable_if<!std::is_arithmetic< T >::value >::type >
	a class for allocating arrays with hip_malloc, specialized for objects More...

class	hip_malloc_deleter
	a deleter for arrays allocated with hip_malloc More...

class	hip_malloc_deleter< T, std::enable_if< std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with hip_malloc, specialized for numbers More...

class	hip_malloc_deleter< T, std::enable_if<!std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with hip_malloc, specialized for objects More...

struct	hip_malloc_uva_allocator
	a class for allocating arrays with hip_malloc_uva More...

struct	hip_malloc_uva_allocator< T, std::enable_if< std::is_arithmetic< T >::value >::type >
	a class for allocating arrays with hip_malloc_uva, specialized for numbers More...

struct	hip_malloc_uva_allocator< T, std::enable_if<!std::is_arithmetic< T >::value >::type >
	a class for allocating arrays with hip_malloc_uva, specialized for objects More...

class	hip_malloc_uva_deleter
	a deleter for arrays allocated with hip_malloc_uva More...

class	hip_malloc_uva_deleter< T, typename std::enable_if< std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with hip_malloc_uva, specialized for numbers More...

class	hip_malloc_uva_deleter< T, typename std::enable_if<!std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with hip_malloc_uva, specialized for objects More...

struct	malloc_allocator
	a class for allocating arrays with malloc More...

struct	malloc_allocator< T, std::enable_if< std::is_arithmetic< T >::value >::type >
	a class for allocating arrays with malloc, specialized for numbers More...

struct	malloc_allocator< T, std::enable_if<!std::is_arithmetic< T >::value >::type >
	a class for allocating arrays with malloc, specialized for objects More...

class	malloc_deleter
	a deleter for arrays allocated with malloc More...

class	malloc_deleter< T, std::enable_if< std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with malloc, specialized for numbers More...

class	malloc_deleter< T, std::enable_if<!std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with malloc, specialized for objects More...

struct	new_allocator
	a class for allocating arrays with new More...

class	new_deleter
	a deleter for arrays allocated with new More...

struct	openmp_allocator
	a class for allocating arrays with OpenMP More...

struct	openmp_allocator< T, std::enable_if< std::is_arithmetic< T >::value >::type >
	a class for allocating arrays with OpenMP, specialized for numbers More...

struct	openmp_allocator< T, std::enable_if<!std::is_arithmetic< T >::value >::type >
	a class for allocating arrays with OpenMP, specialized for objects More...

class	openmp_deleter
	a deleter for arrays allocated with OpenMP More...

class	openmp_deleter< T, std::enable_if< std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with OpenMP, specialized for numbers More...

class	openmp_deleter< T, std::enable_if<!std::is_arithmetic< T >::value >::type >
	a deleter for arrays allocated with OpenMP, specialized for objects More...

class	python_deleter
	a deleter for memory managed from within Python More...

class	stream
	A wrapper around technology specific streams. More...

Typedefs
template<typename T >
using	p_buffer = std::shared_ptr< buffer< T > >
	a shared pointer to an instance of a buffer<T> More...

template<typename T >
using	const_p_buffer = std::shared_ptr< const buffer< T > >
	a shared pointer to an instance of a const buffer<T> More...

Enumerations
enum	buffer_allocator { buffer_allocator::same = -2, buffer_allocator::none = -1, buffer_allocator::cpp = 0, buffer_allocator::malloc = 1, buffer_allocator::cuda = 2, buffer_allocator::cuda_async = 3, buffer_allocator::cuda_uva = 4, buffer_allocator::cuda_host = 5, buffer_allocator::hip = 6, buffer_allocator::hip_uva = 7, buffer_allocator::openmp = 8 }
	allocator types that may be used with hamr::buffer More...

enum	buffer_transfer { buffer_transfer::async = 0, buffer_transfer::sync_cpu = 1, buffer_transfer::sync = 2 }

Functions
const HAMR_EXPORT char *	get_allocator_name (buffer_allocator alloc)
	return the human readable name of the allocator More...

HAMR_EXPORT int	cpu_accessible (buffer_allocator alloc)

HAMR_EXPORT int	cuda_accessible (buffer_allocator alloc)

HAMR_EXPORT int	hip_accessible (buffer_allocator alloc)

HAMR_EXPORT int	openmp_accessible (buffer_allocator alloc)

HAMR_EXPORT void	assert_valid_allocator (buffer_allocator alloc)
	asserts that the passed value is one of the known allocators More...

HAMR_EXPORT buffer_allocator	get_device_allocator ()
	get the allocator type most suitable for the current build configuration. More...

HAMR_EXPORT buffer_allocator	get_host_allocator ()
	get the allocator type most suitable for the current build configuration. More...

template<typename T >
std::ostream &	operator<< (std::ostream &os, const buffer_handle< T > &buf)

template<typename T >
hamr::const_p_buffer< T >	const_ptr (const hamr::p_buffer< T > &v)
	a helper for explicitly casting to a const buffer pointer. More...

template<typename T >
const hamr::buffer< T > &	ref_to (const hamr::const_p_buffer< T > &ptr)
	a helper for getting a reference to pointed to hamr::buffer More...

template<typename T >
hamr::buffer< T > &	ref_to (const hamr::p_buffer< T > &ptr)
	a helper for getting a reference to pointed to hamr::buffer More...

template<typename TT , typename... PP>
auto	get_cpu_accessible (const TT &b, PP &&... args)

template<typename TT , typename... PP>
auto	get_cuda_accessible (const TT &b, PP &&... args)

template<typename TT , typename... PP>
auto	get_hip_accessible (const TT &b, PP &&... args)

template<typename TT , typename... PP>
auto	get_openmp_accessible (const TT &b, PP &&... args)

template<typename TT , typename... PP>
auto	get_device_accessible (const TT &b, PP &&... args)

template<typename... PP>
auto	data (PP &&... args)

template<typename... PP>
auto	pointer (PP &&... args)

template<typename... PP>
void	synchronize (PP &&... args)

template<typename NT >
auto	make_buffer (buffer_allocator alloc, size_t n_elem)

template<typename NT >
auto	make_buffer (buffer_allocator alloc, size_t n_elem, const NT &ival)

int HAMR_EXPORT	get_cuda_device_identifier (int &dev_id)
	gets the device identifier for the first GPU. More...

int HAMR_EXPORT	get_cuda_cpu_identifier (int &dev_id)
	gets the device identifier for the CPU. More...

int HAMR_EXPORT	get_active_cuda_device (int &dev_id)
	gets the currently atcive CUDA device. More...

int HAMR_EXPORT	set_active_cuda_device (int dev_id)
	sets the active CUDA device. returns zero if successful. More...

int HAMR_EXPORT	get_cuda_device (const void *ptr, int &device_id)
	gets the device that owns the given pointer. More...

int HAMR_EXPORT	get_device_identifier (int &dev_id)
	gets the device identifier for the first GPU. More...

int HAMR_EXPORT	get_cpu_identifier (int &dev_id)
	gets the device identifier for the CPU. More...

int HAMR_EXPORT	get_active_device (int &dev_id)
	gets the currently atcive device. More...

int HAMR_EXPORT	set_active_device (int dev_id)
	sets the active device. returns zero if successful. More...

int HAMR_EXPORT	get_device (const void *ptr, int &device_id)
	gets the device that owns the given pointer. More...

constexpr HAMR_EXPORT int	get_verbose ()
	returns the value of the HAMR_VERBOSE environment variable More...

int HAMR_EXPORT	get_hip_device_identifier (int &dev_id)
	gets the device identifier for the first GPU. More...

int HAMR_EXPORT	get_hip_cpu_identifier (int &dev_id)
	gets the device identifier for the CPU. More...

int HAMR_EXPORT	get_active_hip_device (int &dev_id)
	gets the currently atcive HIP device. returns zero if successful. More...

int HAMR_EXPORT	set_active_hip_device (int dev_id)
	sets the active HIP device. returns zero if successful. More...

int HAMR_EXPORT	get_hip_device (const void *ptr, int &device_id)
	gets the device that owns the given pointer. More...

int HAMR_EXPORT	get_openmp_device_identifier (int &dev_id)
	gets the device identifier for the first GPU. More...

int HAMR_EXPORT	get_openmp_cpu_identifier (int &dev_id)
	gets the device identifier for the CPU. More...

int HAMR_EXPORT	get_active_openmp_device (int &dev_id)
	gets the currently atcive HIP device. returns zero if successful. More...

int HAMR_EXPORT	set_active_openmp_device (int dev_id)
	sets the active HIP device. returns zero if successful. More...

int HAMR_EXPORT	get_openmp_device (const void *ptr, int &device_id)
	gets the device that owns the given pointer. More...

Detailed Description

heterogeneous accelerator memory resource

Typedef Documentation

◆ const_p_buffer

template<typename T >

using hamr::const_p_buffer = typedef std::shared_ptr<const buffer<T> >

a shared pointer to an instance of a const buffer<T>

◆ p_buffer

template<typename T >

using hamr::p_buffer = typedef std::shared_ptr<buffer<T> >

a shared pointer to an instance of a buffer<T>

Enumeration Type Documentation

◆ buffer_allocator

enum hamr::buffer_allocator

strong

allocator types that may be used with hamr::buffer

Enumerator
same	propagate the current allocator
none	no allocator specified
cpp	allocates memory with new
malloc	allocates memory with malloc
cuda	allocates memory with cudaMalloc
cuda_async	allocates memory with cudaMallocAsync
cuda_uva	allocates memory with cudaMallocManaged
cuda_host	allocates memory with cudaMallocHost
hip	allocates memory with hipMalloc
hip_uva	allocates memory with hipMallocManaged
openmp	allocates memory with OpenMP device offload API

◆ buffer_transfer

enum hamr::buffer_transfer

strong

flag used to indicate whether or not a transfer operation should be synchronous or not.

Enumerator
async	all operations are asynchronous
sync_cpu	operations moving data from GPU to CPU memory are synchronous
sync	all operations are synchronous

Function Documentation

◆ assert_valid_allocator()

HAMR_EXPORT void hamr::assert_valid_allocator ( buffer_allocator alloc )

inline

asserts that the passed value is one of the known allocators

◆ const_ptr()

template<typename T >

hamr::const_p_buffer<T> hamr::const_ptr ( const hamr::p_buffer< T > & v )

a helper for explicitly casting to a const buffer pointer.

◆ cpu_accessible()

HAMR_EXPORT int hamr::cpu_accessible ( buffer_allocator alloc )

inline

Returns: true if the allocator creates CPU accessible memory

◆ cuda_accessible()

HAMR_EXPORT int hamr::cuda_accessible ( buffer_allocator alloc )

inline

Returns: true if the allocator creates CUDA accessible memory

◆ data()

template<typename... PP>

auto hamr::data ( PP &&... args )

Calls hamr::buffer::data on a number of hamr::buffer instances.

Template Parameters

PP	a paramater pack of hamr::buffer<NT>

Parameters

args	any number of hamr::buffer<NT> instances

Returns: a tuple of NT* one for each hamr::buffer<NT> passed in.

◆ get_active_cuda_device()

int HAMR_EXPORT hamr::get_active_cuda_device ( int & dev_id )

gets the currently atcive CUDA device.

Returns: zero if successful.

◆ get_active_device()

int HAMR_EXPORT hamr::get_active_device ( int & dev_id )

inline

gets the currently atcive device.

Returns: zero if successful.

◆ get_active_hip_device()

int HAMR_EXPORT hamr::get_active_hip_device ( int & dev_id )

gets the currently atcive HIP device. returns zero if successful.

◆ get_active_openmp_device()

int HAMR_EXPORT hamr::get_active_openmp_device ( int & dev_id )

gets the currently atcive HIP device. returns zero if successful.

◆ get_allocator_name()

const HAMR_EXPORT char* hamr::get_allocator_name ( buffer_allocator alloc )

return the human readable name of the allocator

◆ get_cpu_accessible()

template<typename TT , typename... PP>

auto hamr::get_cpu_accessible	(	const TT &	b,
		PP &&...	args
	)

Calls hamr::buffer::get_cpu_accessible on a number of hamr::buffer instances.

Template Parameters

TT	hamr::buffer<NT>
PP	a paramater pack of TT

Parameters

b	a hamr::buffer<NT> instance
args	zero or more hamr::buffer<NT> instances

Returns: a tuple of std::shared_ptr<NT> and NT* one for each hamr::buffer<NT> passed in.

◆ get_cpu_identifier()

int HAMR_EXPORT hamr::get_cpu_identifier ( int & dev_id )

inline

gets the device identifier for the CPU.

Returns: zero if successful.

◆ get_cuda_accessible()

template<typename TT , typename... PP>

auto hamr::get_cuda_accessible	(	const TT &	b,
		PP &&...	args
	)

Calls hamr::buffer::get_cuda_accessible on a number of hamr::buffer instances.

Template Parameters

TT	hamr::buffer<NT>
PP	a paramater pack of TT

Parameters

b	a hamr::buffer<NT> instance
args	zero or more hamr::buffer<NT> instances

Returns: a tuple of std::shared_ptr<NT> and NT* one for each hamr::buffer<NT> passed in.

◆ get_cuda_cpu_identifier()

int HAMR_EXPORT hamr::get_cuda_cpu_identifier ( int & dev_id )

inline

gets the device identifier for the CPU.

Returns: zero if successful.

◆ get_cuda_device()

int HAMR_EXPORT hamr::get_cuda_device	(	const void *	ptr,
		int &	device_id
	)

gets the device that owns the given pointer.

Returns: zero if successful.

◆ get_cuda_device_identifier()

int HAMR_EXPORT hamr::get_cuda_device_identifier ( int & dev_id )

inline

gets the device identifier for the first GPU.

Returns: zero if successful.

◆ get_device()

int HAMR_EXPORT hamr::get_device	(	const void *	ptr,
		int &	device_id
	)

inline

gets the device that owns the given pointer.

Returns: zero if successful.

◆ get_device_accessible()

template<typename TT , typename... PP>

auto hamr::get_device_accessible	(	const TT &	b,
		PP &&...	args
	)

Calls hamr::buffer::get_device_accessible on a number of hamr::buffer instances.

Template Parameters

TT	hamr::buffer<NT>
PP	a paramater pack of TT

Parameters

b	a hamr::buffer<NT> instance
args	zero or more hamr::buffer<NT> instances

Returns: a tuple of std::shared_ptr<NT> and NT* one for each hamr::buffer<NT> passed in.

◆ get_device_allocator()

HAMR_EXPORT buffer_allocator hamr::get_device_allocator ( )

inline

get the allocator type most suitable for the current build configuration.

◆ get_device_identifier()

int HAMR_EXPORT hamr::get_device_identifier ( int & dev_id )

inline

gets the device identifier for the first GPU.

Returns: zero if successful.

◆ get_hip_accessible()

template<typename TT , typename... PP>

auto hamr::get_hip_accessible	(	const TT &	b,
		PP &&...	args
	)

Calls hamr::buffer::get_hip_accessible on a number of hamr::buffer instances.

Template Parameters

TT	hamr::buffer<NT>
PP	a paramater pack of TT

Parameters

b	a hamr::buffer<NT> instance
args	zero or more hamr::buffer<NT> instances

Returns: a tuple of std::shared_ptr<NT> and NT* one for each hamr::buffer<NT> passed in.

◆ get_hip_cpu_identifier()

int HAMR_EXPORT hamr::get_hip_cpu_identifier ( int & dev_id )

inline

gets the device identifier for the CPU.

Returns: zero if successful.

◆ get_hip_device()

int HAMR_EXPORT hamr::get_hip_device	(	const void *	ptr,
		int &	device_id
	)

gets the device that owns the given pointer.

Returns: zero if successful.

◆ get_hip_device_identifier()

int HAMR_EXPORT hamr::get_hip_device_identifier ( int & dev_id )

inline

gets the device identifier for the first GPU.

Returns: zero if successful.

◆ get_host_allocator()

HAMR_EXPORT buffer_allocator hamr::get_host_allocator ( )

inline

get the allocator type most suitable for the current build configuration.

◆ get_launch_props()

HAMR_EXPORT int hamr::get_launch_props	(	int	device_id,
		int *	block_grid_max,
		int &	warp_size,
		int &	max_warps_per_block
	)

query properties for the named CUDA device. retruns non-zero on error

◆ get_openmp_accessible()

template<typename TT , typename... PP>

auto hamr::get_openmp_accessible	(	const TT &	b,
		PP &&...	args
	)

Calls hamr::buffer::get_openmp_accessible on a number of hamr::buffer instances.

Template Parameters

TT	hamr::buffer<NT>
PP	a paramater pack of TT

Parameters

b	a hamr::buffer<NT> instance
args	zero or more hamr::buffer<NT> instances

Returns: a tuple of std::shared_ptr<NT> and NT* one for each hamr::buffer<NT> passed in.

◆ get_openmp_cpu_identifier()

int HAMR_EXPORT hamr::get_openmp_cpu_identifier ( int & dev_id )

gets the device identifier for the CPU.

Returns: zero if successful.

◆ get_openmp_device()

int HAMR_EXPORT hamr::get_openmp_device	(	const void *	ptr,
		int &	device_id
	)

gets the device that owns the given pointer.

Returns: zero if successful.

◆ get_openmp_device_identifier()

int HAMR_EXPORT hamr::get_openmp_device_identifier ( int & dev_id )

inline

gets the device identifier for the first GPU.

Returns: zero if successful.

◆ get_verbose()

constexpr HAMR_EXPORT int hamr::get_verbose ( )

constexpr

returns the value of the HAMR_VERBOSE environment variable

◆ hip_accessible()

HAMR_EXPORT int hamr::hip_accessible ( buffer_allocator alloc )

inline

Returns: true if the allocator creates HIP accessible memory

◆ index_is_valid()

__device__ int hamr::index_is_valid	(	unsigned long	index,
		unsigned long	max_index
	)

inline

bounds check the flat index

◆ make_buffer() [1/2]

template<typename NT >

auto hamr::make_buffer	(	buffer_allocator	alloc,
		size_t	n_elem
	)

constructs an un-initialized hamr::buffer<NT> with space for n_elem allocated and returns it along with the writable pointer to it's contents.

Parameters

[in]	alloc	the allocator to allocate memory with
[in]	n_elem	the initial size of the allocated memory

Returns: a std::tuple with the newly constructed buffer in the first slot and a writable pointer to its internal memory in the second

◆ make_buffer() [2/2]

template<typename NT >

auto hamr::make_buffer	(	buffer_allocator	alloc,
		size_t	n_elem,
		const NT &	ival
	)

constructs an hamr:buffer<NT> with space for n_elem allocated and initialized and returns it along with the writable pointer to it's contents.

Parameters

[in]	alloc	the allocator to allocate memory with
[in]	n_elem	the initial size of the allocated memory
[in]	ival	the value used to initialize the allocated memory

Returns: a std::tuple with the newly constructed buffer in the first slot and a writable pointer to its internal memory in the second

◆ openmp_accessible()

HAMR_EXPORT int hamr::openmp_accessible ( buffer_allocator alloc )

inline

Returns: true if the allocator creates OPENMP accessible memory

◆ partition_thread_blocks() [1/2]

HAMR_EXPORT int hamr::partition_thread_blocks	(	int	device_id,
		size_t	array_size,
		int	warps_per_block,
		dim3 &	block_grid,
		int &	n_blocks,
		dim3 &	thread_grid
	)

Calculate CUDA launch parameters for an arbitrarily large flat array.

Parameters

[in]	device_id	the CUDA device to use. Default values for warps_per_block and block_grid_max are determined by querying the capabilities of the device. If -1 is passed then the currently active device is used.
[in]	array_size	the length of the array being processed
[in]	warps_per_block	number of warps to use per block (your choice). Using a larger number here will result in fewer blocks being processed concurrently.
[out]	block_grid	block dimension kernel launch control
[out]	n_blocks	number of blocks
[out]	thread_grid	thread dimension kernel launch control

Returns: zero if successful and non-zero if an error occurred

◆ partition_thread_blocks() [2/2]

HAMR_EXPORT int hamr::partition_thread_blocks	(	size_t	array_size,
		int	warps_per_block,
		int	warp_size,
		int *	block_grid_max,
		dim3 &	block_grid,
		int &	n_blocks,
		dim3 &	thread_grid
	)

Calculate CUDA launch parameters for an arbitrarily large flat array. See get_launch_props for determining the correct values for warp_size and block_grid_max.

Parameters

[in]	array_size	The length of the array being processed
[in]	warp_size	The number of threads per warp supported on the device
[in]	warps_per_block	The number of warps to use per block (your choice)
[in]	block_grid_max	The maximum number of blocks, in 3-dimensions, supported by the device
[out]	block_grid	The block grid dimension kernel launch control parameter
[out]	n_blocks	The total number of blocks that will be launched
[out]	thread_grid	The thread grid dimension kernel launch control parameter

Returns: zero if successful and non-zero if an error occurred

◆ pointer()

template<typename... PP>

auto hamr::pointer ( PP &&... args )

Calls hamr::buffer::pointer on a number of hamr::buffer instances.

Template Parameters

PP	a paramater pack of hamr::buffer<NT>

Parameters

args	any number of hamr::buffer<NT> instances

Returns: a tuple of std::shared_ptr<NT> one for each hamr::buffer<NT> passed in.

◆ ref_to() [1/2]

template<typename T >

const hamr::buffer<T>& hamr::ref_to ( const hamr::const_p_buffer< T > & ptr )

a helper for getting a reference to pointed to hamr::buffer

◆ ref_to() [2/2]

template<typename T >

hamr::buffer<T>& hamr::ref_to ( const hamr::p_buffer< T > & ptr )

a helper for getting a reference to pointed to hamr::buffer

◆ set_active_cuda_device()

int HAMR_EXPORT hamr::set_active_cuda_device ( int dev_id )

sets the active CUDA device. returns zero if successful.

◆ set_active_device()

int HAMR_EXPORT hamr::set_active_device ( int dev_id )

inline

sets the active device. returns zero if successful.

◆ set_active_hip_device()

int HAMR_EXPORT hamr::set_active_hip_device ( int dev_id )

sets the active HIP device. returns zero if successful.

◆ set_active_openmp_device()

int HAMR_EXPORT hamr::set_active_openmp_device ( int dev_id )

sets the active HIP device. returns zero if successful.

◆ synchronize()

template<typename... PP>

void hamr::synchronize ( PP &&... args )

Calls hamr::buffer::synchronize on a number of hamr::buffer<NT> instances. Note however that one typically need not call synchronize on multiple buffer instances that share the same stream. Synchronizing on any one of them will synchronize all.

Template Parameters

PP	a paramater pack of hamr::buffer<NT>

Parameters

args	any number of hamr::buffer<NT> instances

◆ thread_id_to_array_index()

__device__ unsigned long hamr::thread_id_to_array_index ( )

inline

convert a CUDA index into a flat array index using the partitioning scheme defined in partition_thread_blocks

Classes

Typedefs

Enumerations

Functions

Detailed Description

Typedef Documentation

◆ const_p_buffer

◆ p_buffer

Enumeration Type Documentation

◆ buffer_allocator

◆ buffer_transfer

Function Documentation

◆ assert_valid_allocator()

◆ const_ptr()

◆ cpu_accessible()

◆ cuda_accessible()

◆ data()

◆ get_active_cuda_device()

◆ get_active_device()

◆ get_active_hip_device()

◆ get_active_openmp_device()

◆ get_allocator_name()

◆ get_cpu_accessible()

◆ get_cpu_identifier()

◆ get_cuda_accessible()

◆ get_cuda_cpu_identifier()

◆ get_cuda_device()

◆ get_cuda_device_identifier()

◆ get_device()

◆ get_device_accessible()

◆ get_device_allocator()

◆ get_device_identifier()

◆ get_hip_accessible()

◆ get_hip_cpu_identifier()

◆ get_hip_device()

◆ get_hip_device_identifier()

◆ get_host_allocator()

◆ get_launch_props()

◆ get_openmp_accessible()

◆ get_openmp_cpu_identifier()

◆ get_openmp_device()

◆ get_openmp_device_identifier()

◆ get_verbose()

◆ hip_accessible()

◆ index_is_valid()

◆ make_buffer() [1/2]

◆ make_buffer() [2/2]

◆ openmp_accessible()

◆ partition_thread_blocks() [1/2]

◆ partition_thread_blocks() [2/2]

◆ pointer()

◆ ref_to() [1/2]

◆ ref_to() [2/2]

◆ set_active_cuda_device()

◆ set_active_device()

◆ set_active_hip_device()

◆ set_active_openmp_device()

◆ synchronize()

◆ thread_id_to_array_index()