Classic ML host used internally to calculate getTile and processAllPages requests. More...

#include <mlClassicHost.h>

Public Member Functions
	ClassicHost (Host *host)
	~ClassicHost ()
MLErrorCode	processAllPages (Module &module, MLint outputIndex, SubImageBox region, ProcessAllPagesHandler *handler) ML_RETURN_VALUE_SHOULD_BE_USED
	Internal implementation of processAllPages functionality.
void	shutdownWorkerThreads ()
	Shuts down and frees all worker threads (all threads are joined when this method returns).
void	enableTimeLineProfiling (bool flag)
	Enable time line profiling.
ProcessingTimeLine &	getProcessingTimeLine ()
	Get access to the processing time line.
Computation of image data in different memory allocation variations
MLErrorCode	getTile (PagedImage &outputImage, SubImageBox location, MLDataType dataType, void **data, const ScaleShiftData &scaleShiftData=ScaleShiftData()) ML_RETURN_VALUE_SHOULD_BE_USED
	Requests the image region `location` from the image `outputImage`.
Protected Member Functions
MLErrorCode	_processPagesHelper (PagedImage &outputImage, const SubImageBox &location, MLDataType dataType, void *data, const ScaleShiftData &scaleShiftData, bool calledByProcessAllPages) ML_RETURN_VALUE_SHOULD_BE_USED
	Helper function doing the real page processing work for processAllPages or getTile methods.
MLErrorCode	_getTileCore (PagedImage &outputImage, SubImageBox location, MLDataType dataType, void *data, const ScaleShiftData &scaleShiftData) ML_RETURN_VALUE_SHOULD_BE_USED
	Core of the getTile method: It fills a chunk of memory assuming that it has the correct size.
void	_initializePageData (ThreadDataClass &tData, bool permitInCaching)
	Initializes the buffer `tData` with all settings which is needed to compute one output page.
ThreadDataClass **	_getWorkForWorkerThread (size_t workIndex)
	Returns the work for worker thread with `workIndex`.
void	_increaseErrorCounter ()
	Increases the error counter (threadsafe).
(Sub)Image calculation
void	_calcPages (PagedImage &outputImage, const std::vector< MLint > pageIDs, MLMemoryBlockHandle *memoryBlockHandles)
	For the output image `outputImage`, all invalid pages with numbers given by `pageIDs` are calculated.
Static Protected Member Functions
static void	_calcPagesMT (ThreadDataClass *threadDataContainers, ProcessingTimeLine::TimeLine timeLine)
	Calculates a sequence of output pages where each one has a set of input pages.
Protected Attributes
Parameter
WaitCondition	_workAvailable
	A wait condition to signal waiting worker threads that there is more work to be done.
WaitCondition	_workDone
	A wait condition that signals that all threads have done their work.
Mutex	_workAvailableMutex
	Mutex that protects the work assignment to the worker threads.
std::vector< Thread * >	_workerThreads
	The worker threads that are currently started.
std::vector< ThreadDataClass ** >	_workerThreadClassData
	The offset into the current ThreadDataClass array, sorted by worker threads.
bool	_shutdownWorkerThreadsFlag
	Flag that signals that the worker threads should shutdown.
Mutex	_countSyncMutex
	A mutex used at many position for safe manipulations of members in the host instance by threads.
MLint32	_errorCounter
	Counter for errors detected in calcPagesMT, usually in different threads. Only valid during thread calculations.
MLint32	_numFinishedThreads
	Counter for the number of already finished threads. Only valid during thread calculations.
MLint32	_numLaunchedThreads
	Counter for the number of really launched threads. Only valid during thread calculations.
size_t	_maxNumPagesToCalculateInSingleWorkerThread
	The maximum number of pages that a single worker thread will calculate.
Host *	_host
	The core host.
ProcessingTimeLine	_profile
ProcessingTimeLine::TimeLine *	_guiTimeLine
std::vector < ProcessingTimeLine::TimeLine * >	_workerThreadTimeLines
bool	_useTimeLineProfiling
Friends
struct	CalcPagesWorkerThreadFunction
	make the worker thread function a friend

Detailed Description

Classic ML host used internally to calculate getTile and processAllPages requests.

Definition at line 47 of file mlClassicHost.h.

Constructor & Destructor Documentation

ml::ClassicHost::ClassicHost ( Host * host )

ml::ClassicHost::~ClassicHost ( )

Member Function Documentation

void ml::ClassicHost::_calcPages	(	PagedImage &	outputImage,
		const std::vector< MLint >	pageIDs,
		MLMemoryBlockHandle *	memoryBlockHandles
	)		`[protected]`

For the output image outputImage, all invalid pages with numbers given by pageIDs are calculated.

Errors throw exceptions.

Notes:

An image is composed by many rectangular/box-like pages which are numbered and addressed like voxels (see classes PagedImage).
The Host works page based, i.e., it must know page numbers and PagedImage must not encapsulate pages.
The Host works with untyped and sequential subimages and manages them with SubImage wrappers. These subimages will be composed of pages and passed as module inputs; module outputs will then be a page which again can be part of a subimage passed to the next module input.
A recursion counter is passed for testing purposes and to call certain function after everything has finished.
Errors are thrown as exceptions, either as controlled (allowed) MLErrorCode exceptions or as other exceptions which indicate an uncontrolled worst case crash.

static void ml::ClassicHost::_calcPagesMT	(	ThreadDataClass **	threadDataContainers,
		ProcessingTimeLine::TimeLine *	timeLine
	)		`[static, protected]`

Calculates a sequence of output pages where each one has a set of input pages.

For each page all necessary input/output/buffering information must have been initialized in initializePageData before.

MLErrorCode ml::ClassicHost::_getTileCore	(	PagedImage &	outputImage,
		SubImageBox	location,
		MLDataType	dataType,
		void *	data,
		const ScaleShiftData &	scaleShiftData
	)		`[protected]`

Core of the getTile method: It fills a chunk of memory assuming that it has the correct size.

Updating memory outputs of output images is optionally. Not to be called directly by users!

ThreadDataClass** ml::ClassicHost::_getWorkForWorkerThread ( size_t workIndex ) [protected]

Returns the work for worker thread with workIndex.

void ml::ClassicHost::_increaseErrorCounter ( ) [protected]

Increases the error counter (threadsafe).

void ml::ClassicHost::_initializePageData	(	ThreadDataClass &	tData,
		bool	permitInCaching
	)		`[protected]`

Initializes the buffer tData with all settings which is needed to compute one output page.

permitInCaching must be set true only for one (the last) thread data container if many threads are calculated in parallel. Otherwise more than one thread would modify the memory output of an input.

MLErrorCode ml::ClassicHost::_processPagesHelper	(	PagedImage &	outputImage,
		const SubImageBox &	location,
		MLDataType	dataType,
		void *	data,
		const ScaleShiftData &	scaleShiftData,
		bool	calledByProcessAllPages
	)		`[protected]`

Helper function doing the real page processing work for processAllPages or getTile methods.

If the parameter calledByProcessedAllPages is true: Requests the image region loc from outputIndex of the module. The image data will be stored at memory address data as dataType. The data will be scaled linearly dependent on the settings of the scaleShiftData (see ScaleShiftData).

If the parameter calledByProcessedAllPages is false: Requests the image region location from outputIndex of the module. Values of data, scaleShiftData and dataType are ignored.

void ml::ClassicHost::enableTimeLineProfiling ( bool flag ) [inline]

Enable time line profiling.

Definition at line 94 of file mlClassicHost.h.

ProcessingTimeLine& ml::ClassicHost::getProcessingTimeLine ( ) [inline]

Get access to the processing time line.

Definition at line 97 of file mlClassicHost.h.

MLErrorCode ml::ClassicHost::getTile	(	PagedImage &	outputImage,
		SubImageBox	location,
		MLDataType	dataType,
		void **	data,
		const ScaleShiftData &	scaleShiftData = `ScaleShiftData()`
	)

Requests the image region location from the image outputImage.

If *data is a non NULL pointer, it must be large enough to hold the entire requested image data. If *data is NULL, the getTile function will allocate enough memory for the image data. The image data will be stored in that memory as data type dataType. The data will be scaled linearly dependent on the settings of the scaleShiftData (see ScaleShiftData). The return value data then pointer to the allocated memory. IMPORTANT: Memory allocated with any Host::getTile function must be freed with the Host::freeTile function or with the corresponding mapped freeTile function of the class Module.