Template Class Task

Defined in File Task.hpp

Class Documentation

template<typename CombiDataType = double> class Task

Abstract class for a task that is worked on by a process group (consisting of ProcessGroupWorkers)

In the combination technique, this will correspond to the current solution of the PDE for a component grid of a certain level, along with the operators to advance further in time.

The task’s domain is assumed to be the unit hypercube ((0,1)^d or [0,1)^d [0,1]^d , depending on the boundary conditions)

If the Task uses MPI, it MUST NEVER use MPI_COMM_WORLD, even in calls to linked or included functions. It should instead use the local communicator of the process group: lcomm or theMPISystem()->getLocalComm() .

Public Functions

inline virtual ~Task()

inline DimType getDim() const: get the dimensionality of the task’s domain

inline const LevelVector &getLevelVector() const: get the level vector of the task

inline size_t getID() const: get the unique ID of the task

inline real getCoefficient() const: get the combination coefficient of the task

inline void setID(size_t id)

explicitly set a new ID for the task;

useful if process groups read their task assignment from file instead of receiving it from the manager

make sure ID is continuous and unique!

virtual void run(CommunicatorType lcomm) = 0

run the task

For PDE solvers, this will typically advance the solution to the next timestep

Parameters:: lcomm – local communicator of the process group

inline virtual void changeDir(CommunicatorType lcomm)

change the current working directory

useful for tasks that write output files

Parameters:: lcomm – local communicator of the process group

virtual void init(CommunicatorType lcomm, const std::vector<IndexVector> &decomposition = std::vector<IndexVector>()) = 0

initialize the task

Parameters:

lcomm – local communicator of the process group
decomposition – optional decomposition of the task’s domain

inline bool isFinished() const

flag to indicate that the task has finished

useful for concurrent setups, needs setFinished() to be called at the end of run()

inline void setFinished(bool finished): set the finished flag

inline virtual void getFullGrid(FullGrid<CombiDataType> &fg, RankType lroot, CommunicatorType lcomm, int n = 0)

gathers the entire full grid on a single process

should only be done for small grids!

Parameters:

fg – the full grid to gather to
lroot – rank of the root process
lcomm – local communicator of the process group
n – index of the distributed full grid to gather

virtual DistributedFullGridRef<CombiDataType> getDistributedFullGrid(size_t n = 0) = 0

get the DistributedFullGrid of the task

This is where the high-dimensional data is stored. There may be multiple DistributedFullGrid s, to allow for advanced setups with multple fields (e.g. different species in a plasma simulation, concentration and electric field, etc.)

Parameters:: n – index of the distributed full grid to get

inline virtual ConstDistributedFullGridRef<CombiDataType> getDistributedFullGrid(size_t n) const

template<typename Visitor> inline decltype(auto) visitDistributedFullGrid(Visitor &&visitor, size_t n = 0): visit the DistributedFullGrid with a visitor (convenience wrapper)

template<typename Visitor> inline decltype(auto) visitDistributedFullGrid(Visitor &&visitor, size_t n = 0) const

virtual void setZero() = 0: set all DistributedFullGrid values to zero

inline virtual real getCurrentTimestep() const

get the current time step of the task

useful if there is adaptive timestepping in the solver

inline virtual real getCurrentTime() const

get the current time in the task

useful if there is adaptive timestepping in the solver

inline virtual CombiDataType analyticalSolution(const std::vector<real> &coords, int n = 0) const

get the analytical solution of the task

only applicable where an analytical solution exists, then it can be used for diagnostics such as error integrals

Parameters:: coords – coordinates in the task’s domain

inline virtual std::vector<IndexVector> getDecomposition()

get the decomposition of task’s DistributedFullGrid points to the (Cartesian) process grid

Returns:: vector of index vectors, each containing the lower bounds of the points in one dimension

inline virtual bool isInitialized()

inline virtual void doDiagnostics()

do task-specific postprocessing

useful for tasks that need to do additional work after the main task has been completed

inline virtual void receiveDiagnostics()

receive diagnostics from the task

useful for tasks that need to send additional data to the manager

inline const std::vector<BoundaryType> &getBoundary() const: get the task’s boundary flags

Public Static Functions

static inline void receive(Task **t, RankType source, CommunicatorType comm): receive a task from another rank

static inline void send(const Task *const t, RankType dest, CommunicatorType comm): send a task to another rank

static inline void broadcast(Task **t, RankType root, CommunicatorType comm): broadcast a task to all ranks (typically within the same process group)