stapl::map_reduce_factory< MapWF, ReduceWF, b_coarse > Struct Template Reference

Factory that generates the tasks needed for a map reduce operation. A map reduce operation applies the map work function provided on each set of elements of the input views that have the same offset from the beginning of their respective view. The map operations return values that are combined using the reduce operation provided to form a single result value. The result value is returned to each location in the gang executing the map_reduce call. More...

Classes
struct	result
	Computes the type of the result of the map_reduce PARAGRAPH for the set of views, map operation, and reduce operation provided. More...

Public Member Functions
	map_reduce_factory (MapWF const &map_wf, ReduceWF const &reduce_wf)
coarsener_type	get_coarsener () const
void	reset ()
	Reset all data members related to the number of tasks that have been generated in order to prepare the factory for PARAGRAPH reinvocation.
bool	finished () const
runtime::location_md const &	get_location_md (void) const noexcept
	Returns the location metadata of the location this object registered in.
runtime::location_md &	get_location_md (void) noexcept
	Returns the location metadata of the location this object registered in. noexcept More...
Distributed Object Management
rmi_handle::const_reference const &	get_rmi_handle (void) const noexcept
	Returns the associated rmi_handle.
rmi_handle::reference const &	get_rmi_handle (void) noexcept
	Returns the associated rmi_handle.
size_type	get_location_id (void) const noexcept
	Returns the location id of the local sub-object.
size_type	get_num_locations (void) const noexcept
	Returns the number of locations of the gang of this p_object.
void	advance_epoch (void)
	Advances the epoch of the object. More...
void	unlock (void)
	Used by lock_guard functionality in method of derived classes to request atomicity with respect to incoming RMIs. More...
void	lock (void)
	Used by lock_guard functionality in method of derived classes to request atomicity with respect to incoming RMIs. More...
bool	try_lock (void)
	Attempt to maintain atomicity with respect to incoming RMIs. More...

Public Types
using	coarsener_type = typename std::conditional< b_coarse, default_coarsener, null_coarsener >::type
using	size_type = rmi_handle::size_type

Protected Member Functions
bool	initialized (void) const
void	reset_view_indices ()
	Reset the iterators over the view indices on this location. More...
void	set_view_index_iterator (std::size_t idx, view_index_iterator_base *ptr)
std::size_t	view_indices_size () const
view_index_iterator_base *	get_view_index_iterator (std::size_t n) const

Protected Attributes
bool	m_finished
	Stores whether the derived factory has finished specifying tasks.

Detailed Description

template<typename MapWF, typename ReduceWF, bool b_coarse>
struct stapl::map_reduce_factory< MapWF, ReduceWF, b_coarse >

Factory that generates the tasks needed for a map reduce operation. A map reduce operation applies the map work function provided on each set of elements of the input views that have the same offset from the beginning of their respective view. The map operations return values that are combined using the reduce operation provided to form a single result value. The result value is returned to each location in the gang executing the map_reduce call.

Template Parameters

MapWF	Fine-grain operation to be applied to each set of elements.
ReduceWF	Fine-grain reduce operation used to combine the results of the applications of the map operation.
b_coarse	Indicates whether the map operation provided is coarse-grain.

The factory uses coarse_map_reduce_wf as the work function for the map tasks when the MapWF is a fine-grain operation. The results of the coarse-grain map operations are combined using a binary reduction tree. If the number of tasks used to perform the coarse-grain map operations isn't a power of two a set of binary trees whose widths are the powers of two that sum to equal the number of coarse-grain map operations is generated, and the results of the trees are combined to form the final result. The final result of the computation is pushed to all locations using a set of tasks that form a broadcast tree.

Member Function Documentation

reset_view_indices()

void stapl::task_factory_base::reset_view_indices ( )

protectedinherited

Reset the iterators over the view indices on this location.

This is used in factories where multiple tasks will be generated for each view element to be processed.

get_location_md()

runtime::location_md& stapl::p_object::get_location_md ( void  )

noexceptinherited

Returns the location metadata of the location this object registered in. noexcept

noexcept

advance_epoch()

void stapl::p_object::advance_epoch ( void  )

inherited

Advances the epoch of the object.

Advancing the epoch will flush any pending RMIs. It will also increase the epoch of the current gang if the object is not a named object.

unlock()

void stapl::p_object::unlock ( void  )

inherited

Used by lock_guard functionality in method of derived classes to request atomicity with respect to incoming RMIs.

Mutates counter in the associated runqueue.

lock()

void stapl::p_object::lock ( void  )

inherited

Used by lock_guard functionality in method of derived classes to request atomicity with respect to incoming RMIs.

Mutates counter in the associated runqueue.

try_lock()

bool stapl::p_object::try_lock ( void  )

inherited

Attempt to maintain atomicity with respect to incoming RMIs.

Mutates counter in the associated runqueue.

Returns

true if the lock was able to be acquired. false if it is already locked.

---

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

• map_reduce_prototype.hpp