STAPL: Executable Skeletons

Executable skeletons change functional skeletons from higher-order functions to regular functions by providing access to the input data. More...

Collaboration diagram for Executable Skeletons:

Modules
	Executable Skeletons Internal

Functions
template<typename MapOp , typename ... V>
void	stapl::map_func (MapOp const &map_op, V &&... views)
	Construct and execute a PARAGRAPH that will perform a map operation, applying the fine-grain work function to the element of the views provided. More...

template<typename MapOp , typename ReduceOp , typename ... V>
std::enable_if< !std::is_base_of< skeletons::tags::coarsening_tag, MapOp >::value, stapl::result_of::map_reduce< stapl::use_default, MapOp, ReduceOp, V... >>::type::type	stapl::map_reduce (MapOp const &map_op, ReduceOp const &reduce_op, V &&... v)
	Applies map-reduce on a set of inputs. In the first step the given map operation (zip operation if there is more than one view involved) on the given inputs and the results are reduce using the given reduce operation (`reduce_op`) More...

template<typename Tag , typename MapOp , typename ReduceOp , typename ... V>
stapl::result_of::map_reduce< Tag, MapOp, ReduceOp, V... >::type	stapl::map_reduce (MapOp const &map_op, ReduceOp const &reduce_op, V &&... v)
	Applies map-reduce on the given input by using the specific `map_reduce` implementation specified by the given tag. Current set of tags which can be used for map_reduce are: More...

template<typename Scheduler , typename MapOp , typename ReduceOp , typename ... V>
stapl::result_of::map_reduce< stapl::use_default, MapOp, ReduceOp, V... >::type	stapl::map_reduce_sched (Scheduler const &scheduler, MapOp const &map_op, ReduceOp const &reduce_op, V &&... v)
	Construct and execute a PARAGRAPH that will perform a map operation, applying the fine-grain work function to the element of the views provided. The scheduler provided will be used by the executor processing the tasks of the PARAGRAPH as it is executed. More...

template<typename Scheduler , typename MapOp , typename ... V>
void	stapl::map_func_sched (Scheduler const &scheduler, MapOp const &map_op, V &&... views)
	Construct and execute a PARAGRAPH that will perform a map operation, applying the fine-grain work function to the element of the views provided. The scheduler provided will be used by the executor processing the tasks of the PARAGRAPH as it is executed. More...

template<typename Tag = stapl::use_default, typename View , typename BinaryOp >
std::decay< View >::type::value_type	stapl::reduce (View &&view, BinaryOp const &binop)
	Users can specify which version of the reduce skeleton to be used for reduction algorithm. The possible choices are. More...

template<typename ScanTag , typename CoarseTag , typename View0 , typename View1 , typename Binop >
void	stapl::scan (View0 &&view0, View1 &&view1, Binop binop)
	A scan is an operation in which each element in the results is the result of the cumulative operation on all elements before it. It can be one of two types. More...

template<typename CoarseTag = stapl::use_default, typename View0 , typename View1 , typename Binop >
void	stapl::scan (View0 &&view0, View1 &&view1, Binop binop, bool shift=false)
	A scan is an operation in which each element in the results is the result of the cumulative operation on all elements before it. It can be one of two types. More...

template<typename ScanTag , typename CoarseTag , typename View0 , typename View1 , typename Binop >
std::decay< View0 >::type::value_type	stapl::scan_reduce (View0 &&view0, View1 &&view1, Binop binop)
	combines the functionality of scan and reduce skeleton. More...

template<typename CoarseTag = stapl::use_default, typename View0 , typename View1 , typename Binop >
std::decay< View0 >::type::value_type	stapl::scan_reduce (View0 &&view0, View1 &&view1, Binop binop, bool shift=false)
	Combines the functionality of scan and reduce skeleton. More...

Detailed Description

Executable skeletons change functional skeletons from higher-order functions to regular functions by providing access to the input data.

Function Documentation

◆ map_func()

template<typename MapOp , typename ... V>

void stapl::map_func	(	MapOp const &	map_op,
		V &&...	views
	)

Construct and execute a PARAGRAPH that will perform a map operation, applying the fine-grain work function to the element of the views provided.

Parameters

map_op	Fine-grain map work function.
views	One or more views to process with the map work function.

◆ map_reduce() [1/2]

template<typename MapOp , typename ReduceOp , typename ... V>

std::enable_if< !std::is_base_of<skeletons::tags::coarsening_tag, MapOp>::value, stapl::result_of::map_reduce<stapl::use_default, MapOp, ReduceOp, V...>>::type::type stapl::map_reduce	(	MapOp const &	map_op,
		ReduceOp const &	reduce_op,
		V &&...	v
	)

Applies map-reduce on a set of inputs. In the first step the given map operation (zip operation if there is more than one view involved) on the given inputs and the results are reduce using the given reduce operation (reduce_op)

Parameters

map_op	the operation to be applied in the first step on the given input(s)
reduce_op	the operation to be applied on the results of the the first map phase
v	the inputs

◆ map_reduce() [2/2]

template<typename Tag , typename MapOp , typename ReduceOp , typename ... V>

stapl::result_of::map_reduce<Tag, MapOp, ReduceOp, V...>::type stapl::map_reduce	(	MapOp const &	map_op,
		ReduceOp const &	reduce_op,
		V &&...	v
	)

Applies map-reduce on the given input by using the specific map_reduce implementation specified by the given tag. Current set of tags which can be used for map_reduce are:

stapl::use_default - uses the default coarsened version of map_reduce. This is the most-used version of map_reduce in STAPL.
with_coarsened_wf - in some cases it is desirable to avoid the default workfunction used in the coarsened map phase. One can use this tag to indicate that the given map_op can handle coarse-grain inputs
no_coarsening - when the coarsening phase (data and skeletons) needs to be avoided as a whole.

Parameters

map_op	the operation to be applied in the first step on the given input(s)
reduce_op	the operation to be applied on the results of the the first map phase
v	the inputs

◆ map_reduce_sched()

template<typename Scheduler , typename MapOp , typename ReduceOp , typename ... V>

stapl::result_of::map_reduce<stapl::use_default, MapOp, ReduceOp, V...>::type stapl::map_reduce_sched	(	Scheduler const &	scheduler,
		MapOp const &	map_op,
		ReduceOp const &	reduce_op,
		V &&...	v
	)

Construct and execute a PARAGRAPH that will perform a map operation, applying the fine-grain work function to the element of the views provided. The scheduler provided will be used by the executor processing the tasks of the PARAGRAPH as it is executed.

Parameters

scheduler	Scheduler to employ in the executor processing the PARAGRAPH.
map_op	Fine-grain map work function.
view	One or more views to process with the map work function.

◆ map_func_sched()

template<typename Scheduler , typename MapOp , typename ... V>

void stapl::map_func_sched	(	Scheduler const &	scheduler,
		MapOp const &	map_op,
		V &&...	views
	)

Parameters

scheduler	Scheduler to employ in the executor processing the PARAGRAPH.
map_op	Fine-grain map work function.
views	One or more views to process with the map work function.

◆ reduce()

template<typename Tag = stapl::use_default, typename View , typename BinaryOp >

std::decay<View>::type::value_type stapl::reduce	(	View &&	view,
		BinaryOp const &	binop
	)

Users can specify which version of the reduce skeleton to be used for reduction algorithm. The possible choices are.

stapl::use_default which uses the default coarsened reduction skeleton.
no_coarsening In some cases it is desired to use a reduction in its fine-grained format. One can use this tag to avoid the coarsening phase altogether.

Parameters

view	the input view
binop	the operation used to reduce the values

Template Parameters

Tag	which reduction to be used

Returns: returns the reduction value on each location

See also: algorithm_fwd.hpp for default values.

◆ scan() [1/2]

template<typename ScanTag , typename CoarseTag , typename View0 , typename View1 , typename Binop >

void stapl::scan	(	View0 &&	view0,
		View1 &&	view1,
		Binop	binop
	)

A scan is an operation in which each element in the results is the result of the cumulative operation on all elements before it. It can be one of two types.

exclusive in which the cumulative operation for the value at index i of the output contains all elements before index i in the input
inclusive in which the cumulative operation for element at index i of the output contains all elements before index i and at index i of the input.

The type of the scan to be used is determined by the scan tag passed to this function. In addition, one can choose the coarsening method used for the skeleton by specifying a CoarseTag

Parameters

view0	A one-dimensional view over the input elements that are of a numeric type.
view1	A one-dimensional view over the elements where the result of the scan will be written.
binop	The binary functor that will be used to compute the result of scan operation.

Template Parameters

ScanTag	which scan algorithm to be used
CoarseTag	what type of coarsening to be used (if any)

See also: inclusive_scan; exclusive_scan

◆ scan() [2/2]

template<typename CoarseTag = stapl::use_default, typename View0 , typename View1 , typename Binop >

void stapl::scan	(	View0 &&	view0,
		View1 &&	view1,
		Binop	binop,
		bool	shift = `false`
	)

A scan is an operation in which each element in the results is the result of the cumulative operation on all elements before it. It can be one of two types.

exclusive in which the cumulative operation for the value at index i of the output contains all elements before index i in the input
inclusive in which the cumulative operation for element at index i of the output contains all elements before index i and at index i of the input.

This function specifies the default skeleton to be used for both inclusive and exclusive skeletons. The default case for an exclusive scan is Blelloch and the default for inclusive scan is Binomial.

Parameters

view0	A one-dimensional view over the input elements that are of a numeric type.
view1	A one-dimensional view over the elements where the result of the scan will be written.
binop	The binary functor that will be used to compute the result of scan operation.
shift	Whether to use an exclusive or inclusive scan.

Template Parameters

CoarseTag which type of coarsening to be used for the coarsening of the scan skeleton. One can pass no_coarsening in order to avoid the coarsening as a whole.

See also: inclusive_scan; exclusive_scan

◆ scan_reduce() [1/2]

template<typename ScanTag , typename CoarseTag , typename View0 , typename View1 , typename Binop >

std::decay<View0>::type::value_type stapl::scan_reduce	(	View0 &&	view0,
		View1 &&	view1,
		Binop	binop
	)

combines the functionality of scan and reduce skeleton.

for the scan skeleton it can be one of two types

exclusive in which the cumulative operation for the value at index i of the output contains all elements before index i in the input
inclusive in which the cumulative operation for element at index i of the output contains all elements before index i and at index i of the input.

The type of the scan to be used is determined by the scan tag passed to this function. In addition, one can choose the coarsening method used for the skeleton by specifying a CoarseTag

Parameters

view0	A one-dimensional view over the input elements that are of a numeric type.
view1	A one-dimensional view over the elements where the result of the scan will be written.
binop	The binary functor that will be used to compute the result of scan and reduce operation.

Template Parameters

ScanTag	which scan algorithm to be used
CoarseTag	what type of coarsening to be used (if any)

Returns: the result of reduce skeleton

See also: reduce; scan; inclusive_scan; exclusive_scan

◆ scan_reduce() [2/2]

template<typename CoarseTag = stapl::use_default, typename View0 , typename View1 , typename Binop >

std::decay<View0>::type::value_type stapl::scan_reduce	(	View0 &&	view0,
		View1 &&	view1,
		Binop	binop,
		bool	shift = `false`
	)

Combines the functionality of scan and reduce skeleton.

This function specifies the default skeleton to be used for scan skeleton. The default case for an exclusive scan is Blelloch and the default for inclusive scan is Binomial.

Parameters

view0	A one-dimensional view over the input elements that are of a numeric type.
view1	A one-dimensional view over the elements where the result of the scan will be written.
binop	The binary functor that will be used to compute the result of scan and reduce operation.
shift	Whether to use an exclusive or inclusive scan.

Template Parameters

CoarseTag which type of coarsening to be used for the coarsening of the scan skeleton. One can pass no_coarsening in order to avoid the coarsening as a whole.

Returns: the result of reduce skeleton

See also: reduce; scan; inclusive_scan; exclusive_scan

Modules

Functions

Detailed Description

Function Documentation

◆ map_func()

◆ map_reduce() [1/2]

◆ map_reduce() [2/2]

◆ map_reduce_sched()

◆ map_func_sched()

◆ reduce()

◆ scan() [1/2]

◆ scan() [2/2]

◆ scan_reduce() [1/2]

◆ scan_reduce() [2/2]