GmTaskManager Class Reference

Task manager used for handling parallel executions. More...

#include <gmTaskManager.h>

Collaboration diagram for GmTaskManager:

Public Types
enum	NodePartitionStrategy { NPS_DEFAULT = -1, NPS_SEQUENTIAL_BLOCK = 0, NPS_NUM_STRATEGIES }
	Partition strategies for a parallel node loop. More...
enum	CellPartitionStrategy { CPS_DEFAULT = -1, CPS_SEQUENTIAL_BLOCK = 0, CPS_NUM_STRATEGIES }
	Partition strategies for a parallel cell loop. More...

Public Member Functions
	GmTaskManager (GmSimulationData *simData, GmThreadManager *tm)
	Basic constructor.
GmThreadManager *	threadManager () const
	Returns the associated thread manager.
int	numTasks () const
	Returns the number of tasks added to the queue by addNodetasks or addCellTasks.
template<class T >
GmThreadTaskResult	runParallelNodeLoop (const GmMesh *m, GmAffectedNodes affNodes, T &taskObj, int ntasks=0, NodePartitionStrategy strategy=NPS_DEFAULT, int nworkers=-1, const QVector< int > &taskAffinity=QVector< int >())
	Parallel executes the given task over mesh nodes. More...
template<class T >
void	execParallelNodeLoop (const GmMesh *m, GmAffectedNodes affNodes, T &taskObj, int ntasks=0, NodePartitionStrategy strategy=NPS_DEFAULT, int nworkers=-1, const QVector< int > &taskAffinity=QVector< int >())
	Similar to runParallelNodeLoop() but raising a lua error if the execution was either canceled or aborted.
template<class T >
GmThreadTaskResult	runParallelCellLoop (GmCellMesh *m, bool activeOnly, T &taskObj, int ntasks=0, CellPartitionStrategy strategy=CPS_DEFAULT, int nworkers=-1, const QVector< int > &taskAffinity=QVector< int >())
	Parallel executes the given task over mesh cells. More...
template<class T >
GmThreadTaskResult	runParallelCellLoop (GmCellGroupSet *gs, bool activeOnly, T &taskObj, int ntasks=0, CellPartitionStrategy strategy=CPS_DEFAULT, int nworkers=-1, const QVector< int > &taskAffinity=QVector< int >())
	Similar to runParallelCellLoop(GmCellMesh*, ...) but receiving a CellGroupSet as the cell source instead of the mesh.
template<class T >
void	execParallelCellLoop (GmCellMesh *m, bool activeOnly, T &taskObj, int ntasks=0, CellPartitionStrategy strategy=CPS_DEFAULT, int nworkers=-1, const QVector< int > &taskAffinity=QVector< int >())
	Similar to runParallelCellLoop() but raising a lua error if the execution was either canceled or aborted.
template<class T >
bool	addNodeTasks (const GmMesh *m, GmAffectedNodes affNodes, T &taskObj, int ntasks, NodePartitionStrategy strategy, int nworkers, const QVector< int > &taskAffinity, bool *ok)
	Worker function for adding node tasks to the thread manager. See runParallelNodeLoop() for parameter descriptions. Returns false if no tasks where added to the queue. That can be due to the mesh having no nodes of the specified affNodes or due to an error. The ok parameter is filled with false on errors, true if not. More...
template<class M , class T >
bool	addCellTasks (M *m, bool activeOnly, T &taskObj, int ntasks, CellPartitionStrategy strategy, int nworkers, const QVector< int > &taskAffinity, bool *ok)
	Worker function for adding cell tasks to the thread manager. See runParallelCellLoop() for parameter descriptions Returns false if no tasks where added to the queue. That can be due to the mesh having no (active) cells or due to an error. The ok parameter is filled with false on errors, true if not. More...

Static Public Member Functions
static NodePartitionStrategy	str2NodePartitionStrategy (QString str)
	Converts a string into a NodePartitionStrategy. Unknown strings are converted to 'default'.
static CellPartitionStrategy	str2CellPartitionStrategy (QString str)
	Converts a string into a CellPartitionStrategy. Unknown strings are converted to 'default'.

Private Member Functions
void	adjustExecParameters (int *ntasks, int *nworkers)
	Adjusts task execution parameters, converting possible default options to real values.
QVector< GmNodeTaskDataSet * >	buildNodePartition (const GmMesh *m, int n, GmAffectedNodes affNodes, NodePartitionStrategy strategy) const
	Creates a list with the 'n' parallel node tasks to be executed, given the destination mesh, the desired affected nodes and the partition strategy.
template<class M >
QVector< GmCellTaskDataSet * >	buildCellPartition (M *m, int n, bool activeOnly, CellPartitionStrategy strategy) const
	Creates a list with the 'n' parallel cell tasks to be executed, given the destination mesh, if inactive cells should be considered or not and the partition strategy.

Private Attributes
GmThreadManager *	_threadManager
	The thread manager used to run tasks in multiple threads.
int	_defNumTasks
	The default number of tasks for a parallel loop. Negative numbers are multiples of the number of workers.
NodePartitionStrategy	_defNodeStrategy
	The default strategy for node partitioning.
CellPartitionStrategy	_defCellStrategy
	The default strategy for cell partitioning.

Detailed Description

Task manager used for handling parallel executions.

Member Enumeration Documentation

CellPartitionStrategy

enum GmTaskManager::CellPartitionStrategy

Partition strategies for a parallel cell loop.

Enumerator
CPS_DEFAULT	Uses the default strategy given by simulation options or CPS_SEQUENTIAL_BLOCK if missing.
CPS_SEQUENTIAL_BLOCK	Each worker gets a sequential block of cells.
CPS_NUM_STRATEGIES	The number of partition strategies.

NodePartitionStrategy

enum GmTaskManager::NodePartitionStrategy

Partition strategies for a parallel node loop.

Enumerator
NPS_DEFAULT	Uses the default strategy given by simulation options or NPS_SEQUENTIAL_BLOCK if missing.
NPS_SEQUENTIAL_BLOCK	Each worker gets a sequential block of nodes.
NPS_NUM_STRATEGIES	The number of partition strategies.

Member Function Documentation

addCellTasks()

template<class M , class T >

bool GmTaskManager::addCellTasks	(	M *	m,
		bool	activeOnly,
		T &	taskObj,
		int	ntasks,
		CellPartitionStrategy	strategy,
		int	nworkers,
		const QVector< int > &	taskAffinity,
		bool *	ok
	)

Worker function for adding cell tasks to the thread manager. See runParallelCellLoop() for parameter descriptions Returns false if no tasks where added to the queue. That can be due to the mesh having no (active) cells or due to an error. The ok parameter is filled with false on errors, true if not.

< The user task callable object to be run

< The cell partition assigned to this task

addNodeTasks()

template<class T >

bool GmTaskManager::addNodeTasks	(	const GmMesh *	m,
		GmAffectedNodes	affNodes,
		T &	taskObj,
		int	ntasks,
		NodePartitionStrategy	strategy,
		int	nworkers,
		const QVector< int > &	taskAffinity,
		bool *	ok
	)

Worker function for adding node tasks to the thread manager. See runParallelNodeLoop() for parameter descriptions. Returns false if no tasks where added to the queue. That can be due to the mesh having no nodes of the specified affNodes or due to an error. The ok parameter is filled with false on errors, true if not.

< The user task callable object to be run

< The node partition assigned to this task

runParallelCellLoop()

template<class T >

GmThreadTaskResult GmTaskManager::runParallelCellLoop	(	GmCellMesh *	m,
		bool	activeOnly,
		T &	taskObj,
		int	ntasks = 0,
		CellPartitionStrategy	strategy = CPS_DEFAULT,
		int	nworkers = -1,
		const QVector< int > &	taskAffinity = QVector<int>()
	)

Parallel executes the given task over mesh cells.

Parameters

mesh	The mesh
activeOnly	Should the tasks traverse active cells only?
taskObj	The task to be executed over mesh cell groups. This is a function object (a function pointer or an object with operator()) that will be called in each worker thread receiving as parameters a pointer to its own GmCellTaskDataSet object and a pointer to the const GmThread object in which it is being executed. It should return a GmThreadTaskResult value.
ntasks	The number of cell groups in which the mesh will be partitioned. If less than the number of workers, the used number of workers will be decreased to match. If greater, tasks will be queued and sent to workers as they become available. If 0, the default number of tasks will be used. If > 0, that number will be used. If < 0, the number of tasks will be equal to the effective number of workers multiplied by the absoulte value of the given number of tasks (-1 equal the number of workers, -2 twice the number of workers and so on).
strategy	The startegy that will be used to create cell groups.
nworkers	The number of desired worker threads. If the desired value is greater than the maximum value supported by the thread manager, the maximum value will be used. If -1, the default number of workers configured in the thread manager will be used. A value of 0 means that the tasks will be executed by the main thread.
taskAffinity	An optional vector defining the requested thread id for executing each task. Should be either an empty vector (no thread locking will be used) or a vector with the same number of entries as the number of tasks that will be effectively used in the call. See also the comments on GmThreadManager::addTask()

Returns

Returns GM_THREAD_OK on success (after all the tasks have been completed),

GM_THREAD_CANCELED if the executions has been canceled or GM_THREAD_ABORTED on execution errors.

runParallelNodeLoop()

template<class T >

GmThreadTaskResult GmTaskManager::runParallelNodeLoop	(	const GmMesh *	m,
		GmAffectedNodes	affNodes,
		T &	taskObj,
		int	ntasks = 0,
		NodePartitionStrategy	strategy = NPS_DEFAULT,
		int	nworkers = -1,
		const QVector< int > &	taskAffinity = QVector<int>()
	)

Parallel executes the given task over mesh nodes.

Parameters

mesh	The mesh
affNodes	The type of nodes that should be traversed (geometry, ghost or both)
taskObj	The task to be executed over mesh node groups. This is a function object (a function pointer or an object with operator()) that will be called in each worker thread receiving as parameters a pointer to its own GmNodeTaskDataSet object and a pointer to the const GmThread object

in which it is being executed. It should return a GmThreadTaskResult value.

Parameters

ntasks	The number of node groups in which the mesh will be partitioned. If less than the number of workers, the used number of workers will be decreased to match. If greater, tasks will be queued and sent to workers as they become available. If 0, the default number of tasks will be used. If > 0, that number will be used. If < 0, the number of tasks will be equal to the effective number of workers multiplied by the absoulte value of the given number of tasks (-1 equal the number of workers, -2 twice the number of workers and so on).
strategy	The startegy that will be used to create node groups.
nworkers	The number of desired worker threads. If the desired value is greater than the maximum value supported by the thread manager, the maximum value will be used. If -1, the default number of workers configured in the thread manager will be used. A value of 0 means that the tasks will be executed by the main thread.
taskAffinity	An optional vector defining the requested thread id for executing each task. Should be either an empty vector (no thread locking will be used) or a vector with the same number of entries as the number of tasks that will be effectively used in the call. See also the comments on GmThreadManager::addTask()

Returns

Returns GM_THREAD_OK on success (after all the tasks have been completed),

GM_THREAD_CANCELED if the executions has been canceled or GM_THREAD_ABORTED on execution errors.

---

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

• inc/threads/gmTaskManager.h
• src/threads/gmTaskManager.cpp