Reductions and Scans

Performance-portable reduction and scan operations in AMReX. More...

Classes
struct	amrex::ReduceOpSum
struct	amrex::ReduceOpMin
struct	amrex::ReduceOpMax
struct	amrex::ReduceOpLogicalAnd
struct	amrex::ReduceOpLogicalOr
class	amrex::ReduceData< Ts >
class	amrex::ReduceOps< Ps >
class	amrex::Reducer< Ops, Ts >
	Class for local reductions (e.g., sum, min and max). More...

Functions
template<typename T , typename N , std::enable_if_t< std::is_integral_v< N >, int > = 0>
T	amrex::Reduce::Sum (N n, T const *v, T init_val=0)
	Compute the sum of an array of values.
template<typename T , typename N , typename F , std::enable_if_t< std::is_integral_v< N > &&!std::is_same_v< T *, std::decay_t< F > >, int > = 0>
T	amrex::Reduce::Sum (N n, F const &f, T init_val=0)
	Compute the sum of values generated by a callable.
template<typename T , typename N , std::enable_if_t< std::is_integral_v< N >, int > = 0>
T	amrex::Reduce::Min (N n, T const *v, T init_val=std::numeric_limits< T >::max())
	Compute the minimum of an array of values.
template<typename T , typename N , typename F , std::enable_if_t< std::is_integral_v< N > &&!std::is_same_v< T *, std::decay_t< F > >, int > = 0>
T	amrex::Reduce::Min (N n, F const &f, T init_val=std::numeric_limits< T >::max())
	Compute the minimum of values generated by a callable.
template<typename T , typename N , std::enable_if_t< std::is_integral_v< N >, int > = 0>
T	amrex::Reduce::Max (N n, T const *v, T init_val=std::numeric_limits< T >::lowest())
	Compute the maximum of an array of values.
template<typename T , typename N , typename F , std::enable_if_t< std::is_integral_v< N > &&!std::is_same_v< T *, std::decay_t< F > >, int > = 0>
T	amrex::Reduce::Max (N n, F const &f, T init_val=std::numeric_limits< T >::lowest())
	Compute the maximum of values generated by a callable.
template<typename T , typename N , std::enable_if_t< std::is_integral_v< N >, int > = 0>
std::pair< T, T >	amrex::Reduce::MinMax (N n, T const *v)
	Compute the minimum and maximum of an array of values.
template<typename T , typename N , typename F , std::enable_if_t< std::is_integral_v< N > &&!std::is_same_v< T *, std::decay_t< F > >, int > = 0>
std::pair< T, T >	amrex::Reduce::MinMax (N n, F const &f)
	Compute the minimum and maximum of values generated by a callable.
template<typename T , typename N , typename P , std::enable_if_t< std::is_integral_v< N >, int > = 0>
bool	amrex::Reduce::AnyOf (N n, T const *v, P const &pred)
	Test whether any element in an array satisfies a unary predicate.
template<typename P , int dim>
bool	amrex::Reduce::AnyOf (BoxND< dim > const &box, P const &pred)
	Test whether the predicate is true for any index in a Box.
template<typename T , typename N , typename FIN , typename FOUT , typename TYPE , typename M = std::enable_if_t<std::is_integral_v<N> && (std::is_same_v<std::decay_t<TYPE>,Type::Inclusive> || std::is_same_v<std::decay_t<TYPE>,Type::Exclusive>)>>
T	amrex::Scan::PrefixSum (N n, FIN const &fin, FOUT const &fout, TYPE, RetSum a_ret_sum=retSum)
template<typename N , typename T , typename M = std::enable_if_t<std::is_integral_v<N>>>
T	amrex::Scan::InclusiveSum (N n, T const *in, T *out, RetSum a_ret_sum=retSum)
	Inclusive sum.
template<typename N , typename T , typename M = std::enable_if_t<std::is_integral_v<N>>>
T	amrex::Scan::ExclusiveSum (N n, T const *in, T *out, RetSum a_ret_sum=retSum)
	Exclusive sum.
template<class InIter , class OutIter >
OutIter	amrex::Gpu::inclusive_scan (InIter begin, InIter end, OutIter result)
template<class InIter , class OutIter >
OutIter	amrex::Gpu::exclusive_scan (InIter begin, InIter end, OutIter result)

Detailed Description

Performance-portable reduction and scan operations in AMReX.

These interfaces provide reductions (e.g., sum, min, max) and scan operations (prefix sums) over index spaces and ranges on both CPUs and GPUs. They do not use MPI and operate on data local to the process.

Key functions and classes include:

• amrex::Reducer
• amrex::ReduceOpSum
• amrex::ReduceOpMin
• amrex::ReduceOpMax
• amrex::ReduceOpLogicalAnd
• amrex::ReduceOpLogicalOr
• amrex::ReduceOps
• amrex::ReduceData
• amrex::Reduce::Sum
• amrex::Reduce::Min
• amrex::Reduce::Max
• amrex::Reduce::MinMax
• amrex::Reduce::AnyOf
• amrex::Scan::PrefixSum
• amrex::Scan::InclusiveSum
• amrex::Scan::ExclusiveSum
• amrex::Gpu::inclusive_scan
• amrex::Gpu::exclusive_scan

Function Documentation

AnyOf() [1/2]

template<typename P , int dim>

bool amrex::Reduce::AnyOf	(	BoxND< dim > const &	box,
		P const &	pred
	)

Test whether the predicate is true for any index in a Box.

Note that this is local test that does not involve MPI.

Template Parameters

P	callable type
dim	dimension of the box

Parameters

box	index region for the loop
pred	callable object with the signature of bool(int,int,int).

Returns

the test result.

AnyOf() [2/2]

template<typename T , typename N , typename P , std::enable_if_t< std::is_integral_v< N >, int > = 0>

bool amrex::Reduce::AnyOf	(	N	n,
		T const *	v,
		P const &	pred
	)

Test whether any element in an array satisfies a unary predicate.

Note that this is local test that does not involve MPI.

Template Parameters

T	value type of the elements.
N	integral index type.
P	predicate type

Parameters

n	number of elements to test.
v	pointer to the input data array.
pred	unary predicate object.

Returns

the test result.

exclusive_scan()

template<class InIter , class OutIter >

OutIter amrex::Gpu::exclusive_scan	(	InIter	begin,
		InIter	end,
		OutIter	result
	)

ExclusiveSum()

template<typename N , typename T , typename M = std::enable_if_t<std::is_integral_v<N>>>

T amrex::Scan::ExclusiveSum	(	N	n,
		T const *	in,
		T *	out,
		RetSum	a_ret_sum = retSum
	)

Exclusive sum.

Parameters

n	number of elements
in	input
out	output
a_ret_sum	control the return value

Returns

return the total sum if a_ret_sum is true

inclusive_scan()

template<class InIter , class OutIter >

OutIter amrex::Gpu::inclusive_scan	(	InIter	begin,
		InIter	end,
		OutIter	result
	)

InclusiveSum()

template<typename N , typename T , typename M = std::enable_if_t<std::is_integral_v<N>>>

T amrex::Scan::InclusiveSum	(	N	n,
		T const *	in,
		T *	out,
		RetSum	a_ret_sum = retSum
	)

Inclusive sum.

Parameters

n	number of elements
in	input
out	output
a_ret_sum	control the return value

Returns

return the total sum if a_ret_sum is true

Max() [1/2]

template<typename T , typename N , typename F , std::enable_if_t< std::is_integral_v< N > &&!std::is_same_v< T *, std::decay_t< F > >, int > = 0>

T amrex::Reduce::Max	(	N	n,
		F const &	f,
		T	init_val = std::numeric_limits<T>::lowest()
	)

Compute the maximum of values generated by a callable.

Note that this is local reduction that does not involve MPI.

Template Parameters

T	value type of the reduction.
N	integral index type.
F	callable type with the signature T(N).

Parameters

n	number of elements to reduce.
f	callable object invoked as f(i). For GPU builds, this needs to be a device lambda function or functor.
init_val	initial value (default: std::numeric_limits<T>::lowest()).

Returns

the maximum.

Max() [2/2]

template<typename T , typename N , std::enable_if_t< std::is_integral_v< N >, int > = 0>

T amrex::Reduce::Max	(	N	n,
		T const *	v,
		T	init_val = std::numeric_limits<T>::lowest()
	)

Compute the maximum of an array of values.

Note that this is local reduction that does not involve MPI.

Template Parameters

T	value type of the reduction.
N	integral index type.

Parameters

n	number of elements to reduce.
v	pointer to the input data array.
init_val	initial value (default: std::numeric_limits<T>::lowest()).

Returns

the maximum.

Min() [1/2]

template<typename T , typename N , typename F , std::enable_if_t< std::is_integral_v< N > &&!std::is_same_v< T *, std::decay_t< F > >, int > = 0>

T amrex::Reduce::Min	(	N	n,
		F const &	f,
		T	init_val = std::numeric_limits<T>::max()
	)

Compute the minimum of values generated by a callable.

Note that this is local reduction that does not involve MPI.

Template Parameters

T	value type of the reduction.
N	integral index type.
F	callable type with the signature T(N).

Parameters

n	number of elements to reduce.
f	callable object invoked as f(i). For GPU builds, this needs to be a device lambda function or functor.
init_val	initial value (default: std::numeric_limits<T>::max()).

Returns

the minimum.

Min() [2/2]

template<typename T , typename N , std::enable_if_t< std::is_integral_v< N >, int > = 0>

T amrex::Reduce::Min	(	N	n,
		T const *	v,
		T	init_val = std::numeric_limits<T>::max()
	)

Compute the minimum of an array of values.

Note that this is local reduction that does not involve MPI.

Template Parameters

T	value type of the reduction.
N	integral index type.

Parameters

n	number of elements to reduce.
v	pointer to the input data array.
init_val	initial value (default: std::numeric_limits<T>::max()).

Returns

the minimum.

MinMax() [1/2]

template<typename T , typename N , typename F , std::enable_if_t< std::is_integral_v< N > &&!std::is_same_v< T *, std::decay_t< F > >, int > = 0>

std::pair< T, T > amrex::Reduce::MinMax	(	N	n,
		F const &	f
	)

Compute the minimum and maximum of values generated by a callable.

Note that this is local reduction that does not involve MPI.

Template Parameters

T	value type of the reduction.
N	integral index type.
F	callable type with the signature T(N).

Parameters

n	number of elements to reduce.
f	callable object invoked as f(i). For GPU builds, this needs to be a device lambda function or functor.

Returns

the pair of minimum and maximum.

MinMax() [2/2]

template<typename T , typename N , std::enable_if_t< std::is_integral_v< N >, int > = 0>

std::pair< T, T > amrex::Reduce::MinMax	(	N	n,
		T const *	v
	)

Compute the minimum and maximum of an array of values.

Note that this is local reduction that does not involve MPI.

Template Parameters

T	value type of the reduction.
N	integral index type.

Parameters

n	number of elements to reduce.
v	pointer to the input data array.

Returns

the pair of minimum and maximum.

PrefixSum()

template<typename T , typename N , typename FIN , typename FOUT , typename TYPE , typename M = std::enable_if_t<std::is_integral_v<N> && (std::is_same_v<std::decay_t<TYPE>,Type::Inclusive> || std::is_same_v<std::decay_t<TYPE>,Type::Exclusive>)>>

T amrex::Scan::PrefixSum	(	N	n,
		FIN const &	fin,
		FOUT const &	fout,
		TYPE	,
		RetSum	a_ret_sum = retSum
	)

Sum() [1/2]

template<typename T , typename N , typename F , std::enable_if_t< std::is_integral_v< N > &&!std::is_same_v< T *, std::decay_t< F > >, int > = 0>

T amrex::Reduce::Sum	(	N	n,
		F const &	f,
		T	init_val = 0
	)

Compute the sum of values generated by a callable.

Note that this is local reduction that does not involve MPI.

Template Parameters

T	value type of the reduction.
N	integral index type.
F	callable type with the signature T(N).

Parameters

n	number of elements to reduce.
f	callable object invoked as f(i). For GPU builds, this needs to be a device lambda function or functor.
init_val	initial value (default: 0).

Returns

the sum.

Sum() [2/2]

template<typename T , typename N , std::enable_if_t< std::is_integral_v< N >, int > = 0>

T amrex::Reduce::Sum	(	N	n,
		T const *	v,
		T	init_val = 0
	)

Compute the sum of an array of values.

Note that this is local reduction that does not involve MPI.

Template Parameters

T	value type of the reduction.
N	integral index type.

Parameters

n	number of elements to reduce.
v	pointer to the input data array.
init_val	initial value (default: 0).

Returns

the sum.