Detailed Description

A thread-safe AbstractAligner extension.

The current MASA implementation is not thread-safe. So, if an Aligner executes simultaneous threads, the dispatch methods may generate an inconsistent state leading to segmentation faults. To avoid this behavior, the Aligner must extend the AbstractAlignerSafe class in order to serialize the dispatch calls.

There are two serialization strategies.

Serialize by Mutex: the class uses mutex to lock simultaneous threads. The threads are locked until no other thread is dispatching.
Serialize by Queue: the class enqueue all dispatch executions and a consumer thread dispatches the requisitions to the MASA engine. The threads are only locked during the enqueue process. By now, only the dispatchScore function is queued. To enable this feature, use the createDispatcherQueue and destroyDispatcherQueue in the alignPartition method.

Definition at line 63 of file AbstractAlignerSafe.hpp.

#include <AbstractAlignerSafe.hpp>

Inheritance diagram for AbstractAlignerSafe:

List of all members.

Public Member Functions
	AbstractAlignerSafe ()
virtual	~AbstractAlignerSafe ()
virtual void	dispatchColumn (int j, const cell_t *buffer, int len)
	Delegates to IManager::dispatchColumn()
virtual void	dispatchRow (int i, const cell_t *buffer, int len)
	Delegates to IManager::dispatchRow()
virtual void	dispatchScore (score_t score, int bx=-1, int by=-1)
	Delegates to IManager::dispatchScore()
Protected Member Functions
void	createDispatcherQueue ()
void	destroyDispatcherQueue ()
Private Member Functions
void	executeLoop ()
Static Private Member Functions
static void *	staticFunctionThread (void *arg)
Private Attributes
pthread_t	thread
pthread_mutex_t	mutex
pthread_cond_t	condition
bool	dispatcherQueueActive
queue< dispatch_job_t >	dispatcherQueue

Constructor & Destructor Documentation

AbstractAlignerSafe::AbstractAlignerSafe ( )

Definition at line 27 of file AbstractAlignerSafe.cpp.

AbstractAlignerSafe::~AbstractAlignerSafe ( ) [virtual]

Definition at line 33 of file AbstractAlignerSafe.cpp.

Member Function Documentation

void AbstractAlignerSafe::createDispatcherQueue ( ) [protected]

Definition at line 70 of file AbstractAlignerSafe.cpp.

void AbstractAlignerSafe::destroyDispatcherQueue ( ) [protected]

Definition at line 83 of file AbstractAlignerSafe.cpp.

void AbstractAlignerSafe::dispatchColumn	(	int	j,
		const cell_t *	buffer,
		int	len
	)		`[virtual]`

Delegates to IManager::dispatchColumn()

Notifies to the MASA framework that some cells of a column has been processed.

This function must be called serially for each column. For example, the invocation of dispatchColumn(50000, vector, 100) will dispatch the first 100 cells of the column 50000 to MASA, and the cells are read from the vector[0..99] elements. After this, a call to dispatchColumn(50000, vector, 50) will dispatch the next 50 cells of the same column 5000, and the cells are read from the vector[0..49] elements.

Parameters:

j	the column to be dispatched.
buffer	the vector containing the data (starting from cell 0).
len	the number of cells that will be read from the vector.

See also:: IManager::dispatchColumn()

Reimplemented from AbstractAligner.

Definition at line 37 of file AbstractAlignerSafe.cpp.

void AbstractAlignerSafe::dispatchRow	(	int	i,
		const cell_t *	buffer,
		int	len
	)		`[virtual]`

Delegates to IManager::dispatchRow()

Notifies to the MASA framework that some cells of a row has been processed.

This function must be called serially for each row, analogous to the AbstractAligner::dispatchLastColumn method.

Parameters:

i	the row to be dispatched.
buffer	the vector containing the data (starting from cell 0).
len	the number of cells that will be read from the vector.

See also:: IManager::dispatchRow()

Reimplemented from AbstractAligner.

Definition at line 43 of file AbstractAlignerSafe.cpp.

void AbstractAlignerSafe::dispatchScore	(	score_t	score,
		int	bx = `-1`,
		int	by = `-1`
	)		`[virtual]`

Delegates to IManager::dispatchScore()

Notifies to the MASA framework that a new score has been computed.

This method may be called as many times it is necessary, and the best score will be calculated among all calls of this method.

If the Aligner supports the aligner_capabilities_t::dispatch_block_scores capability, them it must dispatch the score with the bx, by parameters set to the block indices and this method must be called only once for each block.

Parameters:

score	the score to be dispatched
bx	the block position in the horizontal direction, starting from 0 up to AbstractAligner::getGridWidth() minus 1.
by	the block position in the vertical direction, starting from 0 up to AbstractAligner::getGridHeight() minus 1.