stapl::skeletons::skeletons_impl Namespace Reference

This class abstracts the semantics of a reduce to location skeleton by exposing only the necessary information in its representation. More...

Classes
struct	allgather
	This class abstracts the semantics of an allgather skeleton by exposing only the necessary information in its representation. More...
struct	allgather< T, Span, tags::reverse_butterfly< B > >
	This class abstracts the semantics of an allgather skeleton by exposing only the necessary information in its representation. More...
struct	allreduce
	This class abstracts the semantics of an allreduce skeleton by exposing only the necessary information in its representation. More...
struct	allreduce< Op, Span, tags::butterfly< false > >
	This class abstracts the semantics of a butterfly-based allreduce skeleton by exposing only the necessary information in its representation. More...
struct	allreduce< Op, Span, tags::reverse_butterfly< B > >
	This class abstracts the semantics of a reverse-butterfly-based allreduce skeleton by exposing only the necessary information in its representation. More...
struct	alltoall< T, Span, skeletons::tags::butterfly< false > >
	This class abstracts the semantics of an alltoall skeleton by exposing only the necessary information in its representation. More...
struct	alltoall< T, Span, tags::flat >
	This class abstracts the semantics of a flat alltoall skeleton by exposing only the necessary information in its representation. More...
struct	alltoall< T, Span, tags::hybrid >
	This class abstracts the semantics of a hybrid alltoall skeleton by exposing only the necessary information in its representation. More...
struct	alltoall< T, Span, tags::pairwise_exchange >
	This class abstracts the semantics of a hybrid alltoall skeleton by exposing only the necessary information in its representation. More...
struct	alltoall_filter
	The filtering used in hybrid, flat, and pairwise exchange alltoalls for sending only the requested part to other participants. More...
struct	alltoall_filter< T, tags::butterfly< B > >
	A specific filtering used in recursive-doubling (butterfly) implementation of all_to_all. More...
struct	alltoall_merge
	The workfunction used in flat and hybrid alltoall for merging the incoming messages. More...
class	alltoall_merge< T, tags::butterfly< B > >
	Merges the results from two given inputs by putting one chunk of size butterfly_size from the first and then the same size from the second until both inputs are exhausted. More...
struct	alltoall_merge< T, tags::pairwise_exchange >
	The workfunction used in pairwise exchange alltoall for merging the incoming messages. More...
class	alltoall_pd< T, tags::flat >
	In a flat alltoall, parametric dependencies are defined in such a way that each node depends on every node in the producer. More...
class	alltoall_pd< T, tags::hybrid >
	The hybrid alltoall parametric dependency is used for alltoall skeleton in the cases of having large messages. The parametric dependency in this case is defined as a hybrid of the flat and butterfly-based version of alltoall. More...
class	alltoall_pd< T, tags::pairwise_exchange >
	In a pairwise exchange alltoall, parametric dependencies are defined in such a way that each node depends on every node in the producer. More...
struct	binary_tree< Op, Flows, Span, Tag, false >
	This class abstracts the semantics of a non-complete binary tree by exposing only the necessary information in its representation. More...
struct	binary_tree< Op, Flows, Span, Tag, true >
	This class abstracts the semantics of a complete binary tree by exposing only the necessary information in its representation. More...
class	binary_tree_pd
	This binary tree parametric dependency is one of the several possible parametric dependencies that can be used in a binary tree. This parametric dependency is used with tree in order to create reduction trees. This parametric dependency can have different layouts which are determined by the given Tag. More...
class	binomial_tree_pd
	This reduce parametric dependency is one of the several possible parametric dependencies that can be used in reduction. More...
class	bitonic_sort_pd
	The bitonicsort parametric dependency defines the dependencies for each point in the global bitonicsort skeleton. At each local point either a minimum or a maximum will be computed depending on where this point is located in the global skeleton. This will be described in the Op. More...
struct	bitreversal
	This class abstracts the semantics of an bitreversal skeleton by exposing only the necessary information in its representation. More...
class	bitreversal_pd
	A bitreversal parametric dependency used in the bitreversal skeleton. More...
struct	broadcast
	This class abstracts the semantics of a broadcast skeleton by exposing only the necessary information in its representation. More...
struct	butterfly
	This class abstracts the semantics of a butterfly skeleton by exposing only the necessary information in its representation. More...
class	butterfly_pd
	A butterfly parametric dependency defines the dependencies for each element in a big butterfly skeleton. Some algorithms that use butterfly_pd need to know the location of the element in the big butterfly skeleton. An example is DIT FFT algorithm that changes the computation based on the location of the parametric dependency. For the same reason is_position_aware should be specified by user in such cases. In addition, in some cases one would like to read partially from the inputs on a butterfly. That's when the the F (filter) can be used. More...
class	compose
	Compose operator groups the given skeletons together and performs the spawning process in the order of composition (P0, P1, ... Pn). The spawning process starts from the first skeleton (P0) and continues until the last skeleton(Pn) is spawned. More...
struct	compute_return_type
	The return type to be determined for an optimizer by passing the fine-grain type. More...
struct	default_execution_params
	Default execution parameters. More...
struct	default_skeleton_traits
	Default skeleton traits. More...
class	do_while
	Do-while is a composite skeleton that repeats a skeleton for a number of iterations. This number is determined by evaluating the ContinuationP after every iteration. More...
class	do_while_pd
	A do_while parametric dependency represents an element in the dependence graph that controls if the next iteration of a do-while loop should be spawned or not. It does that by wrapping user's ContCond by do_while_stub with and sending memento_stack to it. More...
struct	do_while_stubs
	There are three elements pushed on the top of the memento when this method is invoked. Two of them stand for the current iteration and one is for the one that comes after this iteration. More...
class	elem
	Every parametric dependency is wrapped by this class in order to represent an executable, expandable skeleton. This class has a generic interface for spawning parametric dependencies, and for accessing the flows connected to each node. More...
struct	exclusive_scan< Op, T, tags::blelloch >
	Blelloch scan is an exclusive scan algorithm which is similar to the inclusive scan mentioned in Joseph Jaja's book (. More...
class	execution_params
	Stores the execution parameters to be used for running a a skeleton in STAPL. These parameters are currently: More...
struct	expand_from_pow_two
	This class abstracts the semantics of an expansion skeleton by exposing only the necessary information in its representation. More...
struct	expand_from_pow_two< Op, Span, pos_aware, stapl::use_default, SetResult >
	This class abstracts the semantics of an expansion skeleton by exposing only the necessary information in its representation. More...
class	expand_from_pow_two_pd
	This parametric dependency is used in the postprocessing phase of various algorithms that only accept power-of-two input sizes. More...
struct	fake_flow
	A fake_flow is used for making the result_type computation possible for wrapped skeletons (wrapped_skeleton). More...
struct	fake_flow< mix_view< V, I, CID >, true >
	A fake_flow is used for making the result_type computation possible for wrapped skeletons (wrapped_skeleton). More...
struct	fake_flow< nfast_view< V >, true >
	A fake_flow is used for making the result_type computation possible for wrapped skeletons (wrapped_skeleton). More...
struct	fake_flow< paragraph_impl::paragraph_view< V >, true >
	A fake_flow is used for making the result_type computation possible for wrapped_skeletons. More...
class	farm_init
	Wraps the user provided generator for a farm and passes the farm instance upon invocation to the generator. More...
class	farm_pd
	This parametric dependency is used in farms skeleton to generate the initial seeds of computation. More...
class	farm_stub
	Wraps the environment used in the current spawning process and provides farm functionality to the user workfunctions. More...
struct	fft
	This class abstracts the semantics of an FFT skeleton by exposing only the necessary information in its representation. More...
struct	fft< T, Span, tags::dif >
	This class abstracts the semantics of a Decimation In Frequency (DIF) FFT skeleton by exposing only the necessary information in its representation. More...
struct	fft< T, Span, tags::dit >
	This class abstracts the semantics of a Decimation In Time (DIT) FFT skeleton by exposing only the necessary information in its representation. More...
struct	fft_dif_wf
	This workfunction is used for Decimation In Frequency FFT (DIF) computation. More...
struct	fft_dit_wf
	This workfunction is used for Decimation In Time FFT (DIT) computation. More...
struct	gather
	This class abstracts the semantics of a gather skeleton by exposing only the necessary information in its representation. More...
struct	hand_execute_impl
	Execute a skeleton in place with a set of views. More...
struct	hand_execute_impl< stapl::skeletons::skeletons_impl::zip_reduce< 1, ZipOp, ReduceOp >, View >
	Execute a zip_reduce with one view in place. More...
struct	inclusive_scan< Op, tags::binomial >
	Binomial scan is one of the common implementations of MPI_Scan. More...
struct	inclusive_scan< Op, tags::hillis_steele >
	Hillis Steele scan skeleton is a pointer-jumping based inclusive scan algorithm which has half the height of the scan algorithm in Joseph Jaja's book, but it performs operations and hence is not work optimal. More...
struct	inclusive_scan< Op, tags::jaja >
	This scan skeleton is based on the algorithm given in An Introduction to Parallel Algorithms by Joseph Jaja, page 48. This algorithm's time complexity is . It consists of a reduction tree followed by a modified broadcast skeleton. This scan skeleton is an inclusive scan. More...
struct	is_tgv
	Check if the given view is a paragraph_view. More...
struct	local_entry
	Typed version of local_entry that holds the values that are stored in a local_env. More...
struct	local_entry_base
	The base class for the entities stored in a local_env that can hold heterogeneous types. More...
struct	memento_element
	a memento element is used in memento in order to reduce the compilation time by avoiding boost::bind. More...
struct	memento_element_disposer
	Function object implementing the disposer concept defined by Boost.Intrusive. More...
struct	notify_map
	This class abstracts the semantics of a notify_map skeleton by exposing only the necessary information in its representation. More...
class	notify_map_pd
	A notify_map parametric dependency is similar to a map_pd, however, it waits for NotifCount notifications from the first NotifCount flows. More...
class	odd_even_merge_sort_pd
	The oddevenmergesort parametric dependency defines the dependencies for each point in the global oddevenmergesort skeleton. At each local point either a minimum or a maximum will be computed depending on where this point is located in the global skeleton. This will be described in the Op. More...
class	paragraph_skeleton_manager
	This is a wrapper over the core skeleton_manager that allows the creating of tasks in STAPL task graphs. More...
struct	param_deps_defaults
	Provides the default implementation of methods required for all parametric dependencies in the framework. More...
struct	pointer_jumping
	This class abstracts the semantics of pointer jumping skeleton by exposing only the necessary information in its representation. More...
class	pointer_jumping_pd
	A pointer-jumping parametric dependency can be used in various skeletons. One good example is hillis_steele_scan in which scan is done in a non-work optimal but smaller tree. More...
class	pre_broadcast_pd
	Creates a set_result parametric dependency given a op. Basically this parametric dependency is only used for sinking values of skeletons which couldn't currently set their results directly without using sink_value skeleton. More...
struct	reduce
	This class abstracts the semantics of a reduce skeleton by exposing only the necessary information in its representation. More...
struct	reduce_to_pow_two
	This class abstracts the semantics of a reduce to power-of-two skeleton by exposing only the necessary information in its representation. More...
class	reduce_to_pow_two_pd
	This parametric dependency is used in the preprocessing phase of various algorithms that only accept power-of-two input sizes. More...
class	repeat
	A repeat skeleton spawns a single skeleton for the number of times specified by SizeF. More...
struct	reverse_binary_tree< Op, Flows, Span, Tag, false, Filter, isPositionAware, SetResult >
	This class abstracts the semantics of a non-complete binary tree by exposing only the necessary information in its representation. More...
struct	reverse_binary_tree< Op, Flows, Span, Tag, true, Filter, isPositionAware, SetResult >
	This class abstracts the semantics of a reverse binary tree by exposing only the necessary information in its representation. More...
class	reverse_binary_tree_pd
	A simple broadcast parametric dependency is usually used in combination with reverse_tree to define broadcast and defines the parametric dependency for each element in a broadcast tree. More...
struct	reverse_tree
	This class abstracts the semantics of a reverse k-ary tree by exposing only the necessary information in its representation. More...
struct	scan< Op, tags::scan< Tag, tags::exclusive > >
	Most of the exclusive scan skeletons cannot work directly on inputs with non-power-of-two sizes. As a result, we have to first convert the input to the closest power of two and then perform the scan operation. More...
struct	scan< Op, tags::scan< Tag, tags::inclusive > >
	Most of the inclusive scan skeletons cannot work directly on inputs with non-power-of-two sizes. As a result, we have to first convert the input to the closest power of two and then perform the scan operation. More...
struct	scan_algorithm_tag_type
	Determines the scan algorithm used. More...
class	scan_blelloch_broadcast_pd
	Tis parametric dependency is used in the broadcast tree of Blelloch scan, in which each element applies the operation on inputs from the previous level of the tree and inputs from the reduction tree. Blelloch scan is an exclusive scan algorithm. More...
class	scan_broadcast_pd
	One way to do an inclusive scan is to use the algorithm mentioned in An Introduction to Parallel Algorithms by Joseph Jaja, page 48. This parametric dependency represents the broadcast (top-down) traversal in the above mentioned algorithm. More...
class	scan_expand_from_pow_two_pd
	An parametric dependency used in the postprocessing phase of arbitrary size scan skeletons. More...
class	scan_expand_from_pow_two_pd< Op, tags::scan< Tag, tags::exclusive > >
	A scan expansion parametric dependency used for the postprocessing phase of exclusive scan skeletons. More...
class	scan_expand_from_pow_two_pd< Op, tags::scan< Tag, tags::inclusive > >
	A scan expansion parametric dependency used for the postprocessing phase of inclusive scan skeletons. More...
struct	scatter
	This class abstracts the semantics of a scatter skeleton by exposing only the necessary information in its representation. A scatter skeleton is simply a reverse binary tree with an operation which splits the input into two halves. More...
struct	serial
	This class abstracts the semantics of an serial skeleton by exposing only the necessary information in its representation. More...
class	serial_pd
	A serial_pd parametric dependency used in the serial skeleton. More...
class	set_result_pd
	Creates a set_result parametric dependency given a op. Basically this parametric dependency is only used for sinking values of skeletons which couldn't currently set their results directly without using sink_value skeleton. More...
class	shifted_first_pd
	This parametric dependency is used in the cases similar to the zip_pd, but whenever the reading from the first input flow is shifted by 1. More...
struct	sink_value
	This class abstracts the semantics of a sink_value skeleton by exposing only the necessary information in its representation. More...
class	skeleton_traits< tuple< OptionalPreParams... >, OptionalParams... >
	Skeleton traits define configurable properties of a skeleton. More...
class	spawn_visitor
	spawn_visitor helps in spawning the elements of parametric dependencies by factoring out the task id and number of consumers of a task in this class. More...
class	stencil_2d_pd
	In a 2D stencil, parametric dependencies are defined in such a way that each node depends on every 9 node in its surroundings including itself from the previous timestep. More...
class	tagged_skeleton
	Tags a skeleton with a given Tag. More...
struct	tree
	This class abstracts the semantics of a k-ary tree by exposing only the necessary information in its representation. More...
struct	zip
	This class abstracts the semantics of a filtered zip skeleton by exposing only the necessary information in its representation. More...
class	zip_pd
	A zip parametric dependency is usually used in zip skeleton. It sends input with the same index of each flow to the given Op. More...
struct	zip_reduce
	This class abstracts the semantics of a zip_reduce skeleton by exposing only the necessary information in its representation. More...

Enumerations
enum	do_while_component : std::size_t { body = 0, reduction = 1, continuation = 2 }
	Enum representing the different nested component skeletons in a do_while skeleton. More...

Functions
template<typename ExecutionParams , typename S , typename... Views>
ExecutionParams::result_type	execute (std::false_type, ExecutionParams const &ep, S &&skeleton, Views &&... views)
template<typename ExecutionParams , typename S , typename... Views>
ExecutionParams::result_type	execute (std::true_type, ExecutionParams const &ep, S &&skeleton, Views &&... views)
template<typename OutFlow , typename Coord , std::size_t... Indices>
std::size_t	consumer_count (OutFlow const &out, Coord const &coord, index_sequence< Indices... > &&)

Detailed Description

This class abstracts the semantics of a reduce to location skeleton by exposing only the necessary information in its representation.

A reduce_to_locs skeleton first reduces the input and then broadcasts the result to all locations using a broadcast_to_locs skeleton.

This abstraction not only makes the reconstruction of a reduce_to_locs skeleton easier, but also provides access to the underlying operation in the reduce skeleton. Furthermore, it reduces the symbol size for a reduce_to_locs skeleton, hence, reducing the total compilation time.

Template Parameters

Op	the operation to be used while reducing the input.
Flows	the flow to be used between the reduction and the broadcast_to_locs skeletons.
Span	the iteration space for the elements in each level of the reduction skeleton.
Tag	determines the type of reduction and broadcast to be used.
SetResult	whether the skeleton should set the task results on the pg edge container or not

Enumeration Type Documentation

do_while_component

enum stapl::skeletons::skeletons_impl::do_while_component : std::size_t

strong

Enum representing the different nested component skeletons in a do_while skeleton.

See also

lid_offset

get_nested_skeleton

nested_in_port

nested_out_port