syclcompat::detail Namespace Reference

Atomic extension to implement standard APIs in std::atomic. More...

Classes
struct	IsValidAtomicType
class	dev_mgr
	device manager More...
class	vectorized_binary
class	accessor
class	memory_traits
class	generic_error_type
struct	DataType
struct	DataType< sycl::vec< T, 2 > >

Typedefs
template<typename ValueT >
using	complex_type = detail::complex_namespace::complex< ValueT >

Enumerations
enum	memcpy_direction {   host_to_host , host_to_device , device_to_host , device_to_device ,   automatic }
enum class	pointer_access_attribute { host_only = 0 , device_only , host_device , end }

Functions
static unsigned int	get_tid ()
static fs::path	write_data_to_file (char const *const data, size_t size)
	Write data to temporary file and return absolute path to temporary file. More...
static uint16_t	extract16 (unsigned char const *const ptr)
static uint32_t	extract32 (unsigned char const *const ptr)
static uint64_t	extract64 (unsigned char const *const ptr)
static uint64_t	get_lib_size (char const *const blob)
static kernel_library	load_dl_from_data (char const *const data, size_t size)
template<typename R , typename... Types>
constexpr size_t	getArgumentCount (R(*f)(Types...))
template<int Dim>
sycl::nd_range< 3 >	transform_nd_range (const sycl::nd_range< Dim > &range)
template<auto F, typename... Args>
std::enable_if_t< std::is_invocable_v< decltype(F), Args... >, sycl::event >	launch (const sycl::nd_range< 3 > &range, sycl::queue q, Args... args)
template<auto F, typename... Args>
sycl::event	launch (const sycl::nd_range< 3 > &range, size_t mem_size, sycl::queue q, Args... args)
template<typename ValueT >
ValueT	clamp (ValueT val, ValueT min_val, ValueT max_val)
template<typename ValueT >
int64_t	zero_or_signed_extent (ValueT val, unsigned bit)
	Extend the 'val' to 'bit' size, zero extend for unsigned int and signed extend for signed int. More...
template<typename RetT , bool needSat, typename AT , typename BT , typename BinaryOperation >
constexpr RetT	extend_binary (AT a, BT b, BinaryOperation binary_op)
template<typename RetT , bool needSat, typename AT , typename BT , typename CT , typename BinaryOperation1 , typename BinaryOperation2 >
constexpr RetT	extend_binary (AT a, BT b, CT c, BinaryOperation1 binary_op, BinaryOperation2 second_op)
template<typename ValueT >
bool	isnan (const ValueT a)
static void *	malloc (size_t size, sycl::queue q)
static constexpr size_t	get_pitch (size_t x)
	Calculate pitch (padded length of major dimension x) by rounding up to multiple of 32. More...
static void *	malloc (size_t &pitch, size_t x, size_t y, size_t z, sycl::queue q)
	Malloc pitched 3D data. More...
template<class T >
static sycl::event	fill (sycl::queue q, void *dev_ptr, const T &pattern, size_t count)
	Set pattern to the first count elements of type T starting from dev_ptr. More...
static sycl::event	memset (sycl::queue q, void *dev_ptr, int value, size_t size)
	Set value to the first size bytes starting from dev_ptr in q. More...
template<typename T >
static std::vector< sycl::event >	memset (sycl::queue q, pitched_data data, const T &value, sycl::range< 3 > size)
	Sets value to the 3D memory region pointed by data in q. More...
template<typename T >
static std::vector< sycl::event >	memset (sycl::queue q, void *ptr, size_t pitch, const T &value, size_t x, size_t y)
	Sets val to the pitched 2D memory region pointed by ptr in q. More...
static pointer_access_attribute	get_pointer_attribute (sycl::queue q, const void *ptr)
static memcpy_direction	deduce_memcpy_direction (sycl::queue q, void *to_ptr, const void *from_ptr)
static sycl::event	memcpy (sycl::queue q, void *to_ptr, const void *from_ptr, size_t size, const std::vector< sycl::event > &dep_events={})
static size_t	get_copy_range (sycl::range< 3 > size, size_t slice, size_t pitch)
static size_t	get_offset (sycl::id< 3 > id, size_t slice, size_t pitch)
static std::vector< sycl::event >	memcpy (sycl::queue q, void *to_ptr, const void *from_ptr, sycl::range< 3 > to_range, sycl::range< 3 > from_range, sycl::id< 3 > to_id, sycl::id< 3 > from_id, sycl::range< 3 > size, const std::vector< sycl::event > &dep_events={})
	copy 3D matrix specified by size from 3D matrix specified by from_ptr and from_range to another specified by to_ptr and to_range. More...
static std::vector< sycl::event >	memcpy (sycl::queue q, pitched_data to, sycl::id< 3 > to_id, pitched_data from, sycl::id< 3 > from_id, sycl::range< 3 > size)
	memcpy 2D/3D matrix specified by pitched_data. More...
static std::vector< sycl::event >	memcpy (sycl::queue q, void *to_ptr, const void *from_ptr, size_t to_pitch, size_t from_pitch, size_t x, size_t y)
	memcpy 2D matrix with pitch. More...
static sycl::event	combine_events (std::vector< sycl::event > &events, sycl::queue q)
void	matrix_mem_copy (void *to_ptr, const void *from_ptr, int to_ld, int from_ld, int rows, int cols, int elem_size, sycl::queue queue=syclcompat::get_default_queue(), bool async=false)
template<typename T >
void	matrix_mem_copy (T *to_ptr, const T *from_ptr, int to_ld, int from_ld, int rows, int cols, sycl::queue queue=get_default_queue(), bool async=false)
	Copy matrix data. More...

Variables
auto	exception_handler
	SYCL default exception handler. More...

Detailed Description

Atomic extension to implement standard APIs in std::atomic.

Typedef Documentation

complex_type

template<typename ValueT >

using syclcompat::detail::complex_type = typedef detail::complex_namespace::complex<ValueT>

Definition at line 48 of file math.hpp.

Enumeration Type Documentation

memcpy_direction

enum syclcompat::detail::memcpy_direction

Enumerator
host_to_host
host_to_device
device_to_host
device_to_device
automatic

Definition at line 79 of file memory.hpp.

pointer_access_attribute

enum syclcompat::detail::pointer_access_attribute

strong

Enumerator
host_only
device_only
host_device
end

Definition at line 244 of file memory.hpp.

Function Documentation

clamp()

template<typename ValueT >

ValueT syclcompat::detail::clamp ( ValueT  val, ValueT  min_val, ValueT  max_val  )

inline

Definition at line 51 of file math.hpp.

Referenced by syclcompat::clamp().

combine_events()

static sycl::event syclcompat::detail::combine_events ( std::vector< sycl::event > &  events, sycl::queue  q  )

static

Definition at line 439 of file memory.hpp.

References sycl::_V1::handler::depends_on(), sycl::_V1::handler::host_task(), and sycl::_V1::queue::submit().

Referenced by syclcompat::memcpy_async(), syclcompat::memset_async(), syclcompat::memset_d16_async(), and syclcompat::memset_d32_async().

deduce_memcpy_direction()

static memcpy_direction syclcompat::detail::deduce_memcpy_direction ( sycl::queue  q, void *  to_ptr, const void *  from_ptr  )

static

Definition at line 264 of file memory.hpp.

References device_to_device, device_to_host, end, get_pointer_attribute(), host_to_device, and host_to_host.

extend_binary() [1/2]

template<typename RetT , bool needSat, typename AT , typename BT , typename BinaryOperation >

constexpr RetT syclcompat::detail::extend_binary ( AT  a, BT  b, BinaryOperation  binary_op  )

inlineconstexpr

Definition at line 94 of file math.hpp.

References syclcompat::max(), syclcompat::min(), and zero_or_signed_extent().

extend_binary() [2/2]

template<typename RetT , bool needSat, typename AT , typename BT , typename CT , typename BinaryOperation1 , typename BinaryOperation2 >

constexpr RetT syclcompat::detail::extend_binary ( AT  a, BT  b, CT  c, BinaryOperation1  binary_op, BinaryOperation2  second_op  )

inlineconstexpr

Definition at line 106 of file math.hpp.

References syclcompat::max(), syclcompat::min(), and zero_or_signed_extent().

extract16()

static uint16_t syclcompat::detail::extract16 ( unsigned char const *const  ptr )

inlinestatic

Definition at line 191 of file kernel.hpp.

Referenced by get_lib_size().

extract32()

static uint32_t syclcompat::detail::extract32 ( unsigned char const *const  ptr )

inlinestatic

Definition at line 200 of file kernel.hpp.

Referenced by get_lib_size().

extract64()

static uint64_t syclcompat::detail::extract64 ( unsigned char const *const  ptr )

inlinestatic

Definition at line 211 of file kernel.hpp.

Referenced by get_lib_size().

fill()

template<class T >

static sycl::event syclcompat::detail::fill ( sycl::queue  q, void *  dev_ptr, const T &  pattern, size_t  count  )

inlinestatic

Set pattern to the first count elements of type T starting from dev_ptr.

Template Parameters

T	Datatype of the pattern to be set.

Parameters

q	The queue in which the operation is done.
dev_ptr	Pointer to the device memory address.
pattern	Pattern of type T to be set.
count	Number of elements to be set to the patten.

Returns

An event representing the fill operation.

Definition at line 185 of file memory.hpp.

References sycl::_V1::queue::fill().

get_copy_range()

static size_t syclcompat::detail::get_copy_range ( sycl::range< 3 >  size, size_t  slice, size_t  pitch  )

inlinestatic

Definition at line 292 of file memory.hpp.

References sycl::_V1::detail::array< dimensions >::get().

get_lib_size()

static uint64_t syclcompat::detail::get_lib_size ( char const *const  blob )

inlinestatic

Definition at line 226 of file kernel.hpp.

References extract16(), extract32(), and extract64().

Referenced by syclcompat::load_kernel_library_mem().

get_offset()

static size_t syclcompat::detail::get_offset ( sycl::id< 3 >  id, size_t  slice, size_t  pitch  )

inlinestatic

Definition at line 297 of file memory.hpp.

get_pitch()

static constexpr size_t syclcompat::detail::get_pitch ( size_t  x )

inlinestaticconstexpr

Calculate pitch (padded length of major dimension x) by rounding up to multiple of 32.

Parameters

x	The dimension to be padded (in bytes)

Returns

size_t representing pitched length of dimension x (in bytes).

Definition at line 158 of file memory.hpp.

Referenced by sycl::_V1::image< Dimensions, AllocatorT >::get_pitch(), and malloc().

get_pointer_attribute()

static pointer_access_attribute syclcompat::detail::get_pointer_attribute ( sycl::queue  q, const void *  ptr  )

static

Definition at line 251 of file memory.hpp.

References device_only, sycl::_V1::queue::get_context(), host_device, and host_only.

Referenced by deduce_memcpy_direction().

get_tid()

static unsigned int syclcompat::detail::get_tid ( )

inlinestatic

Definition at line 615 of file device.hpp.

Referenced by syclcompat::detail::dev_mgr::current_device_id(), and syclcompat::detail::dev_mgr::select_device().

getArgumentCount()

template<typename R , typename... Types>

constexpr size_t syclcompat::detail::getArgumentCount ( R(*)(Types...)  f )

constexpr

Definition at line 40 of file launch.hpp.

Referenced by launch().

isnan()

template<typename ValueT >

bool syclcompat::detail::isnan ( const ValueT  a )

inline

Definition at line 121 of file math.hpp.

References syclcompat::isnan().

Referenced by syclcompat::compare(), syclcompat::fmax_nan(), syclcompat::fmin_nan(), syclcompat::isnan(), syclcompat::relu(), and syclcompat::unordered_compare().

launch() [1/2]

template<auto F, typename... Args>

sycl::event syclcompat::detail::launch	(	const sycl::nd_range< 3 > &	range,
		size_t	mem_size,
		sycl::queue	q,
		Args...	args
	)

Definition at line 71 of file launch.hpp.

References getArgumentCount(), syclcompat::local_mem(), sycl::_V1::handler::parallel_for(), and sycl::_V1::queue::submit().

launch() [2/2]

template<auto F, typename... Args>

std::enable_if_t<std::is_invocable_v<decltype(F), Args...>, sycl::event> syclcompat::detail::launch	(	const sycl::nd_range< 3 > &	range,
		sycl::queue	q,
		Args...	args
	)

Definition at line 59 of file launch.hpp.

References getArgumentCount(), and sycl::_V1::queue::parallel_for().

load_dl_from_data()

static kernel_library syclcompat::detail::load_dl_from_data ( char const *const  data, size_t  size  )

inlinestatic

Definition at line 357 of file kernel.hpp.

References write_data_to_file().

Referenced by syclcompat::load_kernel_library(), and syclcompat::load_kernel_library_mem().

malloc() [1/2]

static void* syclcompat::detail::malloc ( size_t &  pitch, size_t  x, size_t  y, size_t  z, sycl::queue  q  )

inlinestatic

Malloc pitched 3D data.

Parameters

[out]	pitch	returns the calculated pitch (in bytes)
[in]	x	width of the allocation (in bytes)
[in]	y	height of the allocation
[in]	z	depth of the allocation
[in]	q	The queue in which the operation is done.

Returns

A pointer to the allocated memory

Definition at line 169 of file memory.hpp.

References get_pitch(), and malloc().

malloc() [2/2]

static void* syclcompat::detail::malloc ( size_t  size, sycl::queue  q  )

inlinestatic

Definition at line 150 of file memory.hpp.

References sycl::_V1::queue::get_context(), and sycl::_V1::queue::get_device().

Referenced by malloc().

matrix_mem_copy() [1/2]

template<typename T >

void syclcompat::detail::matrix_mem_copy ( T *  to_ptr, const T *  from_ptr, int  to_ld, int  from_ld, int  rows, int  cols, sycl::queue  queue = get_default_queue(), bool  async = false  )

inline

Copy matrix data.

The default leading dimension is column.

Parameters

[out]	to_ptr	A pointer points to the destination location.
[in]	from_ptr	A pointer points to the source location.
[in]	to_ld	The leading dimension the destination matrix.
[in]	from_ld	The leading dimension the source matrix.
[in]	rows	The number of rows of the source matrix.
[in]	cols	The number of columns of the source matrix.
[in]	queue	The queue where the routine should be executed.
[in]	async	If this argument is true, the return of the function does NOT guarantee the copy is completed.

Definition at line 122 of file util.hpp.

References matrix_mem_copy().

matrix_mem_copy() [2/2]

void syclcompat::detail::matrix_mem_copy ( void *  to_ptr, const void *  from_ptr, int  to_ld, int  from_ld, int  rows, int  cols, int  elem_size, sycl::queue  queue = syclcompat::get_default_queue(), bool  async = false  )

inline

Definition at line 86 of file util.hpp.

References memcpy(), sycl::_V1::event::wait(), and sycl::_V1::ext::intel::experimental::esimd::wait().

Referenced by matrix_mem_copy().

memcpy() [1/4]

static std::vector<sycl::event> syclcompat::detail::memcpy ( sycl::queue  q, pitched_data  to, sycl::id< 3 >  to_id, pitched_data  from, sycl::id< 3 >  from_id, sycl::range< 3 >  size  )

inlinestatic

memcpy 2D/3D matrix specified by pitched_data.

Definition at line 420 of file memory.hpp.

References syclcompat::pitched_data::get_data_ptr(), syclcompat::pitched_data::get_pitch(), syclcompat::pitched_data::get_y(), and syclcompat::memcpy().

memcpy() [2/4]

static sycl::event syclcompat::detail::memcpy ( sycl::queue  q, void *  to_ptr, const void *  from_ptr, size_t  size, const std::vector< sycl::event > &  dep_events = {}  )

static

Definition at line 283 of file memory.hpp.

Referenced by matrix_mem_copy().

memcpy() [3/4]

static std::vector<sycl::event> syclcompat::detail::memcpy ( sycl::queue  q, void *  to_ptr, const void *  from_ptr, size_t  to_pitch, size_t  from_pitch, size_t  x, size_t  y  )

inlinestatic

memcpy 2D matrix with pitch.

Definition at line 430 of file memory.hpp.

References syclcompat::memcpy().

memcpy() [4/4]

static std::vector<sycl::event> syclcompat::detail::memcpy ( sycl::queue  q, void *  to_ptr, const void *  from_ptr, sycl::range< 3 >  to_range, sycl::range< 3 >  from_range, sycl::id< 3 >  to_id, sycl::id< 3 >  from_id, sycl::range< 3 >  size, const std::vector< sycl::event > &  dep_events = {}  )

inlinestatic

copy 3D matrix specified by size from 3D matrix specified by from_ptr and from_range to another specified by to_ptr and to_range.

Definition at line 304 of file memory.hpp.

memset() [1/3]

template<typename T >

static std::vector<sycl::event> syclcompat::detail::memset ( sycl::queue  q, pitched_data  data, const T &  value, sycl::range< 3 >  size  )

inlinestatic

Sets value to the 3D memory region pointed by data in q.

Template Parameters

T	The type of the element to be set.

Parameters

[in]	q	The queue in which the operation is done.
[in]	data	Pointer to the pitched device memory region.
[in]	value	The value to be set.
[in]	size	3D memory region by number of elements.

Returns

An event list representing the memset operations.

Definition at line 211 of file memory.hpp.

References sycl::_V1::detail::array< dimensions >::get(), syclcompat::pitched_data::get_data_ptr(), syclcompat::pitched_data::get_pitch(), and syclcompat::pitched_data::get_y().

memset() [2/3]

static sycl::event syclcompat::detail::memset ( sycl::queue  q, void *  dev_ptr, int  value, size_t  size  )

inlinestatic

Set value to the first size bytes starting from dev_ptr in q.

Parameters

q	The queue in which the operation is done.
dev_ptr	Pointer to the device memory address.
value	Value to be set.
size	Number of bytes to be set to the value.

Returns

An event representing the memset operation.

Definition at line 197 of file memory.hpp.

References sycl::_V1::queue::memset().

Referenced by memset().

memset() [3/3]

template<typename T >

static std::vector<sycl::event> syclcompat::detail::memset ( sycl::queue  q, void *  ptr, size_t  pitch, const T &  value, size_t  x, size_t  y  )

inlinestatic

Sets val to the pitched 2D memory region pointed by ptr in q.

Template Parameters

T	The type of the element to be set.

Parameters

[in]	q	The queue in which the operation is done.
[in]	ptr	Pointer to the virtual device memory.
[in]	pitch	The pitch size by number of elements, including padding.
[in]	value	The value to be set.
[in]	x	The width of memory region by number of elements.
[in]	y	The height of memory region by number of elements.

Returns

An event list representing the memset operations.

Definition at line 237 of file memory.hpp.

References memset().

transform_nd_range()

template<int Dim>

sycl::nd_range<3> syclcompat::detail::transform_nd_range

(

const sycl::nd_range< Dim > &

range

)

Definition at line 45 of file launch.hpp.

References sycl::_V1::nd_range< Dimensions >::get_global_range(), and sycl::_V1::nd_range< Dimensions >::get_local_range().

write_data_to_file()

static fs::path syclcompat::detail::write_data_to_file ( char const *const  data, size_t  size  )

inlinestatic

Write data to temporary file and return absolute path to temporary file.

Temporary file is created in a temporary directory both of which have random names with only the user having access permissions. Only one temporary file will be created in the temporary directory.

Definition at line 96 of file kernel.hpp.

Referenced by load_dl_from_data().

zero_or_signed_extent()

template<typename ValueT >

int64_t syclcompat::detail::zero_or_signed_extent ( ValueT  val, unsigned  bit  )

inline

Extend the 'val' to 'bit' size, zero extend for unsigned int and signed extend for signed int.

Definition at line 85 of file math.hpp.

Referenced by extend_binary().

Variable Documentation

exception_handler

auto syclcompat::detail::exception_handler

inline

Initial value:

= [](sycl::exception_list exceptions) {
  for (std::exception_ptr const &e : exceptions) {
    try {
      std::rethrow_exception(e);
    } catch (sycl::exception const &e) {
      std::cerr << "[SYCLcompat] Caught asynchronous SYCL exception:"
                << std::endl
                << e.what() << std::endl
                << "Exception caught at file:" << __FILE__
                << ", line:" << __LINE__ << std::endl;
    }
  }
}

SYCL default exception handler.

Definition at line 64 of file device.hpp.