hmlp::mpitree::DistSplitTask< NODE > Class Template Reference

This is the default ball tree splitter. Given coordinates, compute the direction from the two most far away points. Project all points to this line and split into two groups using a median select. More...

#include <tree_mpi.hpp>

Inheritance diagram for hmlp::mpitree::DistSplitTask< NODE >:

Public Member Functions
void	Set (NODE *user_arg)
void	DependencyAnalysis ()
void	Execute (Worker *user_worker)
Public Member Functions inherited from hmlp::Task
	Task ()
	(Default) Task constructor. More...
	~Task ()
	(Default) Task destructor. More...
TaskStatus	GetStatus ()
	Status is a private member.
void	SetStatus (TaskStatus status)
	Move foward to the next status.
void	SetBatchStatus (TaskStatus status)
void	Submit ()
	Ask the runtime to create an normal task in file.
virtual void	Set (string user_name, void(*user_function)(Task *), void *user_arg)
	This is only for virtual function pointer.
virtual void	Prefetch (Worker *)
void	Enqueue ()
void	Enqueue (size_t tid)
bool	TryEnqueue ()
	Try to dispatch the task if there is no dependency left.
void	ForceEnqueue (size_t tid)
void	CallBackWhileWaiting ()
	This is the callback function for the owner of thenested task. More...
virtual void	GetEventRecord ()
void	DependenciesUpdate ()
	Update the my outgoing and children's incoming edges. More...
void	Acquire ()
void	Release ()
bool	IsNested ()

Public Attributes
NODE *	arg = NULL
Public Attributes inherited from hmlp::Task
class Worker *	worker = NULL
string	name
string	label
int	taskid
float	cost = 0
bool	priority = false
Event	event
void(*	function )(Task *)
void *	arg
volatile int	n_dependencies_remaining = 0
deque< Task * >	in
deque< Task * >	out
Lock *	task_lock = NULL
Task *	next = NULL
volatile bool	stealable = true
volatile int	created_by = 0

Detailed Description

template<typename NODE>
class hmlp::mpitree::DistSplitTask< NODE >

This is the default ball tree splitter. Given coordinates, compute the direction from the two most far away points. Project all points to this line and split into two groups using a median select.

Need to explit the parallelism.Parallel median search TODO: Can be parallelized

Member Function Documentation

template<typename NODE>

void hmlp::mpitree::DistSplitTask< NODE >::Set ( NODE *  user_arg )

inline

Setup the event

Asuume computation bound

"HIGH" priority

---

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

• /home/chenhany/Projects/hmlp/gofmm/tree_mpi.hpp