concurrent reduce implementation More...

#include <SimdTransformReduce.hpp>

Classes
struct	ScalarReducer

Public Member Functions
constexpr	SimdTransformReduce ()=default

Protected Member Functions
template<uint32_t T_maxConcurrencyInByte, alpaka::concepts::Alignment T_MemAlignment = AutoAligned>
ALPAKA_FN_INLINE ALPAKA_FN_ACC constexpr auto	transformReduce (concepts::Acc auto const &acc, alpaka::concepts::Vector auto extents, auto const &neutralElement, auto &&reduceFunc, auto &&func, alpaka::concepts::IDataSource auto &&data0, alpaka::concepts::IDataSource auto &&... dataN) const

Private Member Functions
constexpr auto const &	asParent () const
ALPAKA_FN_INLINE constexpr auto	getReducer (onAcc::concepts::Acc auto const &, auto &&reduceOp) const
	Get the reducer functor.
ALPAKA_FN_INLINE constexpr auto	getReducer (onAcc::concepts::Acc auto const &acc, auto &&reduceOp) const
template<uint32_t T_simdWidth, uint32_t T_numSimdPerFnCall, alpaka::concepts::Alignment T_MemAlignment>
ALPAKA_FN_INLINE ALPAKA_FN_ACC constexpr auto	reduceSimdPackExecution (auto const &acc, alpaka::concepts::Vector auto numElements, auto const &neutralElement, auto &&userReduceFunc, auto &&func, alpaka::concepts::IDataSource auto &&data0, alpaka::concepts::IDataSource auto &&... dataN) const

Static Private Member Functions
template<alpaka::concepts::Alignment T_MemAlignment, uint32_t T_width>
static ALPAKA_FN_INLINE constexpr auto	executeDoTransform (concepts::Acc auto const &acc, auto const &dataIdx, auto &&func, alpaka::concepts::IDataSource auto &&... data)
template<alpaka::concepts::Alignment T_MemAlignment, uint32_t T_width, uint32_t... T_repeat>
static ALPAKA_FN_INLINE constexpr auto	executeReduce (concepts::Acc auto const &acc, auto &iter, std::integer_sequence< uint32_t, T_repeat... >, auto &&reduceFunc, auto &&func, alpaka::concepts::IDataSource auto &&... data)
	Calls the transform functor T_repeat times and reduces the results with the given reduce function.
template<alpaka::concepts::Alignment T_MemAlignment, uint32_t T_width, uint32_t... T_repeat>
static ALPAKA_FN_INLINE constexpr void	executeReduceInto (concepts::Acc auto const &acc, auto &iter, std::integer_sequence< uint32_t, T_repeat... >, auto &result, auto &&reduceFn, auto &&transformFn, alpaka::concepts::IDataSource auto &&... data)
	Reduce simdized packs into a single simdized pack with the given reduce function.
template<uint32_t... T_idx>
ALPAKA_FN_INLINE static ALPAKA_FN_ACC constexpr auto	loadAncExecuteScalarOp (std::integer_sequence< uint32_t, T_idx... >, auto &&op, auto const &acc, auto &&func, auto &&... data)
template<uint32_t... T_repeat>
static ALPAKA_FN_INLINE constexpr auto	makeAdvanceIterators (auto &iter, std::integer_sequence< uint32_t, T_repeat... >)
	advance the iterator T_repeat times
template<uint32_t T_simdWidth, uint32_t T_numSimdPerFnCall, alpaka::concepts::Alignment T_MemAlignment>
static ALPAKA_FN_INLINE constexpr void	reduceNextSimdized (auto const &acc, auto &iter, auto &tmpReturn, auto &&reduceFn, auto &&transformFn, alpaka::concepts::IDataSource auto &&data0, alpaka::concepts::IDataSource auto &&... dataN)
	Reduce T_numSimdPerFnCall simdized packs.
ALPAKA_FN_INLINE static ALPAKA_FN_ACC constexpr decltype(auto)	wrapTransformFunc (auto &&transformFunc)

Detailed Description

template<typename T_Parent>
struct alpaka::onAcc::internal::SimdTransformReduce< T_Parent >

concurrent reduce implementation

Definition at line 28 of file SimdTransformReduce.hpp.

Constructor & Destructor Documentation

◆ SimdTransformReduce()

template<typename T_Parent>

alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::SimdTransformReduce ( )

constexprdefault

Member Function Documentation

◆ asParent()

template<typename T_Parent>

auto const & alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::asParent ( ) const

inlineconstexprprivate

Definition at line 354 of file SimdTransformReduce.hpp.

◆ executeDoTransform()

template<typename T_Parent>

template<alpaka::concepts::Alignment T_MemAlignment, uint32_t T_width>

ALPAKA_FN_INLINE constexpr auto alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::executeDoTransform	(	concepts::Acc auto const &	acc,
		auto const &	dataIdx,
		auto &&	func,
		alpaka::concepts::IDataSource auto &&...	data )

inlinestaticconstexprprivate

Definition at line 146 of file SimdTransformReduce.hpp.

◆ executeReduce()

template<typename T_Parent>

template<alpaka::concepts::Alignment T_MemAlignment, uint32_t T_width, uint32_t... T_repeat>

ALPAKA_FN_INLINE constexpr auto alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::executeReduce	(	concepts::Acc auto const &	acc,
		auto &	iter,
		std::integer_sequence< uint32_t, T_repeat... >	,
		auto &&	reduceFunc,
		auto &&	func,
		alpaka::concepts::IDataSource auto &&...	data )

inlinestaticconstexprprivate

Calls the transform functor T_repeat times and reduces the results with the given reduce function.

The calls to the functor are independent and compile time unrolled to support instruction parallelism. In contrast to executeReduceInto() the register footprint is larger because T_repeat temporary results will be holt. This allows the compiler to use instruction level parallelism. Call this function if result of reduceFunc is a SIMD pack.

Parameters

iter	the caller must ensure tha the interator can be increased T_repeat times without jumping over iter.end()

Returns: a single simdized pack

Definition at line 184 of file SimdTransformReduce.hpp.

◆ executeReduceInto()

template<typename T_Parent>

template<alpaka::concepts::Alignment T_MemAlignment, uint32_t T_width, uint32_t... T_repeat>

ALPAKA_FN_INLINE constexpr void alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::executeReduceInto	(	concepts::Acc auto const &	acc,
		auto &	iter,
		std::integer_sequence< uint32_t, T_repeat... >	,
		auto &	result,
		auto &&	reduceFn,
		auto &&	transformFn,
		alpaka::concepts::IDataSource auto &&...	data )

inlinestaticconstexprprivate

Reduce simdized packs into a single simdized pack with the given reduce function.

In contrast to executeReduce() the register footprint is lower because all intermediate results are directly reduced into the result variable. Call this function if the type of result is a simdized pack is not a SIMD pack.

Parameters

result The results of reduceFn with the result of transformFn will be reduced into this simdized pack.

Definition at line 223 of file SimdTransformReduce.hpp.

◆ getReducer() [1/2]

template<typename T_Parent>

ALPAKA_FN_INLINE constexpr auto alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::getReducer	(	onAcc::concepts::Acc auto const &	,
		auto &&	reduceOp ) const

inlineconstexprprivate

Get the reducer functor.

Returns: wrapped functor in case the input is

See also: ScalarFunc else the identity

Definition at line 342 of file SimdTransformReduce.hpp.

◆ getReducer() [2/2]

template<typename T_Parent>

ALPAKA_FN_INLINE constexpr auto alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::getReducer	(	onAcc::concepts::Acc auto const &	acc,
		auto &&	reduceOp ) const

inlineconstexprprivate

Definition at line 348 of file SimdTransformReduce.hpp.

◆ loadAncExecuteScalarOp()

template<typename T_Parent>

template<uint32_t... T_idx>

ALPAKA_FN_INLINE static ALPAKA_FN_ACC constexpr auto alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::loadAncExecuteScalarOp	(	std::integer_sequence< uint32_t, T_idx... >	,
		auto &&	op,
		auto const &	acc,
		auto &&	func,
		auto &&...	data )

inlinestaticconstexprprivate

Definition at line 100 of file SimdTransformReduce.hpp.

◆ makeAdvanceIterators()

template<typename T_Parent>

template<uint32_t... T_repeat>

ALPAKA_FN_INLINE constexpr auto alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::makeAdvanceIterators	(	auto &	iter,
		std::integer_sequence< uint32_t, T_repeat... >	)

inlinestaticconstexprprivate

advance the iterator T_repeat times

We do not check if the iterator points to a valid element, the caller must ensure that we can safely advance the iterator T_repeat time without jumping over iter.end().

Template Parameters

T_repeat Number of time sthe iterator should be advanced.

Returns: Tuple with T_repeat times iterators.

Definition at line 164 of file SimdTransformReduce.hpp.

◆ reduceNextSimdized()

template<typename T_Parent>

template<uint32_t T_simdWidth, uint32_t T_numSimdPerFnCall, alpaka::concepts::Alignment T_MemAlignment>

ALPAKA_FN_INLINE constexpr void alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::reduceNextSimdized	(	auto const &	acc,
		auto &	iter,
		auto &	tmpReturn,
		auto &&	reduceFn,
		auto &&	transformFn,
		alpaka::concepts::IDataSource auto &&	data0,
		alpaka::concepts::IDataSource auto &&...	dataN )

inlinestaticconstexprprivate

Reduce T_numSimdPerFnCall simdized packs.

Definition at line 248 of file SimdTransformReduce.hpp.

◆ reduceSimdPackExecution()

template<typename T_Parent>

template<uint32_t T_simdWidth, uint32_t T_numSimdPerFnCall, alpaka::concepts::Alignment T_MemAlignment>

ALPAKA_FN_INLINE ALPAKA_FN_ACC constexpr auto alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::reduceSimdPackExecution	(	auto const &	acc,
		alpaka::concepts::Vector auto	numElements,
		auto const &	neutralElement,
		auto &&	userReduceFunc,
		auto &&	func,
		alpaka::concepts::IDataSource auto &&	data0,
		alpaka::concepts::IDataSource auto &&...	dataN ) const

inlineconstexprprivate

To avoid an overflow in the index range we divide first by the number of elements per function call and then by the number of workers.

Definition at line 360 of file SimdTransformReduce.hpp.

◆ transformReduce()

template<typename T_Parent>

template<uint32_t T_maxConcurrencyInByte, alpaka::concepts::Alignment T_MemAlignment = AutoAligned>

ALPAKA_FN_INLINE ALPAKA_FN_ACC constexpr auto alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::transformReduce	(	concepts::Acc auto const &	acc,
		alpaka::concepts::Vector auto	extents,
		auto const &	neutralElement,
		auto &&	reduceFunc,
		auto &&	func,
		alpaka::concepts::IDataSource auto &&	data0,
		alpaka::concepts::IDataSource auto &&...	dataN ) const

inlineconstexprprotected

Definition at line 34 of file SimdTransformReduce.hpp.

◆ wrapTransformFunc()

template<typename T_Parent>

ALPAKA_FN_INLINE static ALPAKA_FN_ACC constexpr decltype(auto) alpaka::onAcc::internal::SimdTransformReduce< T_Parent >::wrapTransformFunc ( auto && transformFunc )

inlinestaticconstexprprivate

Definition at line 110 of file SimdTransformReduce.hpp.

The documentation for this struct was generated from the following file:

/home/docs/checkouts/readthedocs.org/user_builds/alpaka3/checkouts/latest/include/alpaka/onAcc/internal/SimdTransformReduce.hpp

Classes

Public Member Functions

Protected Member Functions

Private Member Functions

Static Private Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ SimdTransformReduce()

Member Function Documentation

◆ asParent()

◆ executeDoTransform()

◆ executeReduce()

◆ executeReduceInto()

◆ getReducer() [1/2]

◆ getReducer() [2/2]

◆ loadAncExecuteScalarOp()

◆ makeAdvanceIterators()

◆ reduceNextSimdized()

◆ reduceSimdPackExecution()

◆ transformReduce()

◆ wrapTransformFunc()