Kernels

Local concurrent operations. More...

Classes
struct	full_range
	This class computes full range size if prefix or subinterval of the range is not specified. More...
struct	prefix_range
	This class computes the range based on the prefix specified. More...
struct	sub_range
	This class computes subinterval of a range based on the starting and ending indices provided. More...

Macros
#define	forall(it, P, name)
	A parallel range-for loop.
#define	reduceall(it, ref, p, R, T, name)
	A parallel reduction loop.

Functions
template<class M , class... RR>
auto	mdiota_view (const M &m, RR... rr)
	Compute the Cartesian product of several intervals of integers.
template<typename Policy , typename Lambda >
void	parallel_for (Policy &&p, Lambda &&lambda, const std::string &name="")
	Call a function on each element of a range, potentially in parallel.
template<class R , class T , typename Policy , typename Lambda >
T	parallel_reduce (Policy &&p, Lambda &&lambda, const std::string &name="")
	Perform a reduction based on the elements of a range, potentially in parallel.

Detailed Description

Local concurrent operations.

If Kokkos is not available, they simply execute serially. To avoid unnecessary copies, one needs to pass a view since the ranges provided by the user are copied.

Macro Definition Documentation

forall

#define forall	(		it,
			P,
			name
	)

A parallel range-for loop.

Follow with a compound statement and ;. Often the elements of range (and thus the values of it) are indices for other ranges.

Parameters

it	variable name to introduce
P	sized random-access range
name	debugging name, convertible to std::string

reduceall

#define reduceall	(		it,
			ref,
			p,
			R,
			T,
			name
	)

A parallel reduction loop.

Follow with a compound statement to form an expression. Often the elements of range (and thus the values of it) are indices for other ranges.

Parameters

it	variable name to introduce for elements
ref	variable name to introduce for storing results; call it with each value participating in the reduction
p	sized random-access range
R	reduction operation type
T	data type
name	debugging name, convertible to std::string

Returns

the reduced result

Function Documentation

mdiota_view()

template<class M , class... RR>

auto mdiota_view	(	const M &	m,
		RR...	rr
	)

Compute the Cartesian product of several intervals of integers.

Parameters

m	mdspan or mdcolex object
rr	full_range, prefix_range, or sub_range objects for each dimension

Returns

sized random-access range of std::array objects, each with one index of type range_index for each argument in rr

parallel_for()

template<typename Policy , typename Lambda >

void parallel_for	(	Policy &&	p,
		Lambda &&	lambda,
		const std::string &	name = ""
	)

Call a function on each element of a range, potentially in parallel.

If GPU support is available, lambda is executed there.

Parameters

p	sized random-access range
name	operation name, for debugging

parallel_reduce()

template<class R , class T , typename Policy , typename Lambda >

T parallel_reduce	(	Policy &&	p,
		Lambda &&	lambda,
		const std::string &	name = ""
	)

Perform a reduction based on the elements of a range, potentially in parallel.

If GPU support is available, lambda is executed there.

Template Parameters

R	reduction operation type
T	data type
Lambda	function of an element of p and a function object that calls the latter with each value participating in the reduction

Parameters

p	sized random-access range
name	operation name, for debugging