ThreadSchedulerMutexes : Version of a ThreadSchedulerLargestCost that also makes sure to not try to schedule two tasks with the same mutex at the same time. More...

#include <ThreadSchedulerMutexes.h>

Inheritance diagram for Mantid::Kernel::ThreadSchedulerMutexes:

Public Member Functions
void	clear () override final
	Empty out the queue.

bool	empty () override

void	finished (Task *task, size_t threadnum) override
	Signal to the scheduler that a task is complete.

std::shared_ptr< Task >	pop (size_t threadnum) override
	Retrieves the next Task to execute.

void	push (std::shared_ptr< Task > newTask) override
	Add a Task to the queue.

size_t	size () override
	Returns the size of the queue.

	ThreadSchedulerMutexes ()=default

	~ThreadSchedulerMutexes () override

Public Member Functions inherited from Mantid::Kernel::ThreadScheduler
virtual void	abort (std::runtime_error exception)
	Signal to the scheduler that a task is complete.

bool	getAborted ()
	Returns true if the execution was aborted.

std::runtime_error	getAbortException ()
	Returns the exception that was caught, if any.

bool	hasBusyThreads () const
	Returns true if any thread is busy.

void	markBusy ()
	Mark a thread as busy.

void	markIdle ()
	Mark a thread as no longer busy. See markBusy().

	ThreadScheduler ()
	Constructor.

double	totalCost ()
	Returns the total cost of all Task's in the queue.

double	totalCostExecuted ()
	Returns the total cost of all Task's in the queue.

virtual	~ThreadScheduler ()=default
	Destructor.

Protected Types
using	InnerMap = std::multimap< double, std::shared_ptr< Task > >
	Map to tasks, sorted by cost.

using	SuperMap = std::map< std::shared_ptr< std::mutex >, InnerMap >
	Map to maps, sorted by Mutex*.

Protected Attributes
std::set< std::shared_ptr< std::mutex > >	m_mutexes
	Vector of currently used mutexes.

SuperMap	m_supermap
	A super map; first key = a Mutex * Inside it: second key = the cost.

Protected Attributes inherited from Mantid::Kernel::ThreadScheduler
bool	m_aborted
	The run was aborted due to an exception.

std::runtime_error	m_abortException
	The exception that aborted the run.

double	m_cost
	Total cost of all tasks.

double	m_costExecuted
	Accumulated cost of tasks that have been executed (popped)

std::atomic< size_t >	m_numBusyThreads {0}
	Number of threads that are getting or running a task.

std::mutex	m_queueLock
	Mutex to prevent simultaneous access to the queue.

Detailed Description

ThreadSchedulerMutexes : Version of a ThreadSchedulerLargestCost that also makes sure to not try to schedule two tasks with the same mutex at the same time.

This scheduler also sorts by largest cost so as to optimize allocation that way.

NOTE: The performance of "pop"ping a task is much slower if you have a very large number of different mutexes; this scheduler is better suited if you only have a few (e.g. one for DiskIO, and NULL for calculations). Popping a task scales with the N^2 where N is the number of different mutexes.

Author: Janik Zikovsky

Date: 2011-02-25 16:39:43.233991

Definition at line 37 of file ThreadSchedulerMutexes.h.

Member Typedef Documentation

◆ InnerMap

using Mantid::Kernel::ThreadSchedulerMutexes::InnerMap = std::multimap<double, std::shared_ptr<Task> >

protected

Map to tasks, sorted by cost.

Definition at line 166 of file ThreadSchedulerMutexes.h.

◆ SuperMap

using Mantid::Kernel::ThreadSchedulerMutexes::SuperMap = std::map<std::shared_ptr<std::mutex>, InnerMap>

protected

Map to maps, sorted by Mutex*.

Definition at line 168 of file ThreadSchedulerMutexes.h.

Constructor & Destructor Documentation

◆ ThreadSchedulerMutexes()

Mantid::Kernel::ThreadSchedulerMutexes::ThreadSchedulerMutexes ( )

default

◆ ~ThreadSchedulerMutexes()

Mantid::Kernel::ThreadSchedulerMutexes::~ThreadSchedulerMutexes ( )

inlineoverride

Definition at line 41 of file ThreadSchedulerMutexes.h.

References clear().

Member Function Documentation

◆ clear()

void Mantid::Kernel::ThreadSchedulerMutexes::clear ( )

inlinefinaloverridevirtual

Empty out the queue.

Implements Mantid::Kernel::ThreadScheduler.

Definition at line 151 of file ThreadSchedulerMutexes.h.

References Mantid::Kernel::ThreadScheduler::m_cost, Mantid::Kernel::ThreadScheduler::m_costExecuted, Mantid::Kernel::ThreadScheduler::m_queueLock, and m_supermap.

Referenced by ~ThreadSchedulerMutexes().

◆ empty()

bool Mantid::Kernel::ThreadSchedulerMutexes::empty ( )

inlineoverridevirtual

Returns: true if the queue is empty

Implements Mantid::Kernel::ThreadScheduler.

Definition at line 140 of file ThreadSchedulerMutexes.h.

References Mantid::Kernel::ThreadScheduler::m_queueLock, and m_supermap.

◆ finished()

void Mantid::Kernel::ThreadSchedulerMutexes::finished	(	Task *	task,
		size_t	threadnum
	)

inlineoverridevirtual

Signal to the scheduler that a task is complete.

Parameters

task	:: the Task that was completed.
threadnum	:: unused argument

Reimplemented from Mantid::Kernel::ThreadScheduler.

Definition at line 117 of file ThreadSchedulerMutexes.h.

References Mantid::Kernel::Task::getMutex(), m_mutexes, Mantid::Kernel::ThreadScheduler::m_queueLock, and UNUSED_ARG.

◆ pop()

std::shared_ptr< Task > Mantid::Kernel::ThreadSchedulerMutexes::pop ( size_t threadnum )

inlineoverridevirtual

Retrieves the next Task to execute.

Parameters

threadnum :: ID of the calling thread.

Returns: a Task pointer to execute.

Implements Mantid::Kernel::ThreadScheduler.

Definition at line 54 of file ThreadSchedulerMutexes.h.

References m_mutexes, Mantid::Kernel::ThreadScheduler::m_queueLock, m_supermap, and UNUSED_ARG.

◆ push()

void Mantid::Kernel::ThreadSchedulerMutexes::push ( std::shared_ptr< Task > newTask )

inlineoverridevirtual

Add a Task to the queue.

Parameters

newTask :: Task to add to queue

Implements Mantid::Kernel::ThreadScheduler.

Definition at line 44 of file ThreadSchedulerMutexes.h.

References Mantid::Kernel::ThreadScheduler::m_cost, Mantid::Kernel::ThreadScheduler::m_queueLock, and m_supermap.

◆ size()

size_t Mantid::Kernel::ThreadSchedulerMutexes::size ( )

inlineoverridevirtual

Returns the size of the queue.

Implements Mantid::Kernel::ThreadScheduler.

Definition at line 128 of file ThreadSchedulerMutexes.h.

References Mantid::Kernel::ThreadScheduler::m_queueLock, and m_supermap.

Member Data Documentation

◆ m_mutexes

std::set<std::shared_ptr<std::mutex> > Mantid::Kernel::ThreadSchedulerMutexes::m_mutexes

protected

Vector of currently used mutexes.

Definition at line 175 of file ThreadSchedulerMutexes.h.

Referenced by finished(), and pop().

◆ m_supermap

SuperMap Mantid::Kernel::ThreadSchedulerMutexes::m_supermap

protected

A super map; first key = a Mutex * Inside it: second key = the cost.

Definition at line 172 of file ThreadSchedulerMutexes.h.

Referenced by clear(), empty(), pop(), push(), and size().

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

Framework/Kernel/inc/MantidKernel/ThreadSchedulerMutexes.h

Public Member Functions

Protected Types

Protected Attributes

Detailed Description

Member Typedef Documentation

◆ InnerMap

◆ SuperMap

Constructor & Destructor Documentation

◆ ThreadSchedulerMutexes()

◆ ~ThreadSchedulerMutexes()

Member Function Documentation

◆ clear()

◆ empty()

◆ finished()

◆ pop()

◆ push()

◆ size()

Member Data Documentation

◆ m_mutexes

◆ m_supermap