peano4 Namespace Reference

Namespaces
namespace	dastgen2
namespace	datamanagement
namespace	datamodel
namespace	grid
	The grid namespace is Peano's core.
namespace	maps
	The map namespaces hosts various map implementations.
namespace	output
namespace	parallel
namespace	Project
namespace	runner
namespace	solversteps
namespace	stacks
namespace	toolbox
namespace	utils
namespace	visualisation

Data Structures
struct	SplitInstruction
	Instruction to split. More...

Typedefs
typedef std::map< int, SplitInstruction >	SplitSpecification

Functions
void	writeCopyrightMessage ()
	You can invoke this operation manually, but it will also implicitly be triggered by the init routines.
void	fillLookupTables ()
	Fill Lookup Tables.
int	initParallelEnvironment (int *argc, char ***argv)
	Init Parallel Environment.
void	shutdownParallelEnvironment ()
	Shutdown all the parallel environment, i.e.
void	initSingletons (const tarch::la::Vector< Dimensions, double > &offset, const tarch::la::Vector< Dimensions, double > &width, const std::bitset< Dimensions > &periodicBC=0)
	Fire up all the singletons.
void	shutdownSingletons ()
	The very first thing I have to do is to shut down Node.
tarch::tests::TestCase *	getUnitTests ()
	Construct suite of unit tests for Peano 4.

Typedef Documentation

SplitSpecification

typedef std::map<int, SplitInstruction> peano4::SplitSpecification

Definition at line 44 of file grid.h.

Function Documentation

fillLookupTables()

void peano4::fillLookupTables

(

)

Fill Lookup Tables.

Fill all the lookup tables used within the application. As lookup tables are used by many operations, I suggest to call this operation as soon as possible.

There shall no error occur in this operation. Thus, it does not return any code.

getUnitTests()

tarch::tests::TestCase * peano4::getUnitTests

(

)

Construct suite of unit tests for Peano 4.

Please destroy after usage. If you want to add a new unit test (set) for Peano, you have to add it here. In principle, the order in this routine should not make a difference: Eventually, all tests should pass anyway.

initParallelEnvironment()

int peano4::initParallelEnvironment	(	int *	argc,
		char ***	argv )

Init Parallel Environment.

Inits the parallel environment. If the parallel mode is not set, the operation detoriates to nop. The function returns 0 if everything is o.k., it returns -2 otherwise. Please call this operation before you call any other operation that could result in an error. I suggest to call it right after fillLookupTables().

Please note that Peano 4 considers both shared memory and distributed memory to be a parallel environment.

init might change the variables passed. If you want to parse the command line arguments, use the values returned. If you use the arguments directly without calling initParallelEnvironment() they might contain MPI values not important to the program.

Default parameters

The routine calls peano4::parallel::Node::initMPIDatatypes(), but it also sets a default timeout of 60s. You can alter this default by calling tarch::mpi::Rank::setDeadlockTimeOut(). All Python projects have a feature Project.set_mpi_timeout() which you can use to alter the default timeout from the Python side.

Usage/implementation details

You may not use the trace macros before this operation has invoked the init operation. Otherwise, the getRank() assertion fails, as the node has not been configured correctly.

Invoke with an address operator before that.

peano4::initParallelEnvironment(&argc,&argv);

Context within Peano

Calling initParallelEnvironment() has to be done as one of the very first things, i.e. before you init the logging, or run tests, or ... However, please note that it is really about the static context of the code. A central object within any MPI-parallel Peano 4 application is the singleton peano4::parallel::Node which you have to initialise (and shutdown) separately.

initSingletons()

void peano4::initSingletons	(	const tarch::la::Vector< Dimensions, double > &	offset,
		const tarch::la::Vector< Dimensions, double > &	width,
		const std::bitset< Dimensions > &	periodicBC = 0 )

Fire up all the singletons.

Singletons that I don't touch are:

• tarch::mpi::Rank, as the rank is not a service and is handled separately by the initParallelEnvironment().

Note that there is a bug when passing in initializer lists to this function. If the number of elements in "offset" is greater than Dimensions, then "width" will borrow some elements that were intended for "offset". Best practice is to initialize vectors that are to be passed in before the call to initSingletons. That way the compiler will throw an error if the sizes are not correct.

shutdownParallelEnvironment()

void peano4::shutdownParallelEnvironment

(

)

Shutdown all the parallel environment, i.e.

free all MPI datatypes and close down MPI. This also turns off the shared memory environment. Before this happens, you have to shutdown the node such that everybody knows that we are going down. So you have to call Node::shutdown() before you trigger this operation. This is your responsibility.

The routine first adds a barrier. This barrier is necessary. If the very last activity of all ranks is for example to plot stuff, they typically use global semaphores as well. To make these semaphores work, we still require that all nodes call receiveDanglingMessages(). It is only after everyone has done their dump, that we can shut down the shared memory system. This is the reason the barrier has to come after the node's shutdown and then has to be a Peano 4 barrier which still invokes receiveDanglingMessages() on all services.

Once all shared memory tasks have terminated, we free the MPI datatypes.

Eventually, we shut down the MPI rank.

Once this routine has terminated, do not add any barrier() anymore!

See also

peano4::parallel::Node::shutdown()

shutdownSingletons()

void peano4::shutdownSingletons

(

)

The very first thing I have to do is to shut down Node.

This shutdown will tell all the other ranks to go down as well.

writeCopyrightMessage()

void peano4::writeCopyrightMessage

(

)

You can invoke this operation manually, but it will also implicitly be triggered by the init routines.