tarch Namespace Reference

Have to include this header, as I need access to the SYCL_EXTERNAL keyword. More...

Namespaces
namespace	accelerator
namespace	internal
namespace	la
	My collection of tiny vector operations.
namespace	logging
namespace	mpi
namespace	multicore
	This page describes Peano 4's multithreading namespace.
namespace	plotter
namespace	services
namespace	tests
namespace	timing

Data Structures
class	Enumerator
	Simple rank-global enumerator. More...

Enumerations
enum class	MemoryLocation { Heap , ManagedSharedAcceleratorDeviceMemory }
enum class	MemoryUsageFormat { MByte }

Functions
void	gpuAbort ()
	Delegates to std::abort() if no GPU offloading is active.
std::string	toString (MemoryLocation value)
template<class T = double>
T *	allocateMemory (std::size_t count, MemoryLocation location, int device=accelerator::Device::HostDevice)
	Allocates memory on the specified memory location.
void	freeMemory (void *data, MemoryLocation location, int device=accelerator::Device::HostDevice)
	Free memory.
std::size_t	padSizeToAlignment (std::size_t size, std::size_t alignment)
double *	memset (double *dest, double ch, size_t byteCount)
	Alternative GPU-ready version of memset.
void	shutdownNonCriticalAssertionEnvironment ()
	peano4::shutdownParallelEnvironment().
void	initNonCriticalAssertionEnvironment ()
	Register the assertion tag from the global communicator.
void	enableNonCriticalAssertions (bool value)
	Switch noncritical assertions on/off.
void	triggerNonCriticalAssertion (std::string file, int line, std::string expression, std::string parameterValuePairs)
	Trigger a non-critical assertion.
bool	hasNonCriticalAssertionBeenViolated ()
int	getMemoryUsage (MemoryUsageFormat format)
	Method for getting the application's memory footprint.
int	getTotalMemory (MemoryUsageFormat format)
int	getFreeMemory (MemoryUsageFormat format)
tarch::tests::TestCase *	getUnitTests ()
	Please destroy after usage.

Detailed Description

Have to include this header, as I need access to the SYCL_EXTERNAL keyword.

tarch denotes the Technical Architectures.

It is used within Peano, but it is designed in a way that it is more or less independent of Peano, i.e. Peano uses the tarch but not the other way round.

Please note that the tarch in the Peano4 repository/branch is not the same as the tarch in previous Peano versions. I've cleaned it up quite a lot and old features that are not used anymore are removed.

level

• PeanoDebug = 1: Enable tracing
• PeanoDebug = 2: Enable assertions (includes level 1)
• PeanoDebug = 4: Enable debug messages (includes level 2)

libraries

By default, Peano 4 builds all core libraries in three different versions:

• A debug variant which holds a compile with PeanoDebug=4
• A trace variant which holds a compile with PeanoDebug=1
• A production variants with PeanoDebug=0

Note: The trace variant does not contain any assertions, as assertions are typically expensive and slow down the code considerably.

Enumeration Type Documentation

MemoryLocation

enum class tarch::MemoryLocation

strong

Enumerator
Heap	Create data on the heap of the local device. That is, if you run on an accelerator, use the accelerator's heap. If you invoke it on the host, use the host's accelerator. The alignment of this routine can be controlled via the compile time constant AlignmentOnHeap. It is typically set (to a default) within the corresponding compiler-specific header-file in tarch/compiler.
ManagedSharedAcceleratorDeviceMemory	To be used on host only.

Definition at line 58 of file accelerator.h.

MemoryUsageFormat

enum class tarch::MemoryUsageFormat

strong

Enumerator
MByte

Definition at line 35 of file tarch.h.

Function Documentation

allocateMemory()

template<class T = double>

T * tarch::allocateMemory	(	std::size_t	count,
		MemoryLocation	location,
		int	device = accelerator::Device::HostDevice )

Allocates memory on the specified memory location.

The device id is an optional argument that is required for memory allocation on an accelerator. Default is -1 (HostDevice). For memory allocation on accelerators, passing a device id >= 0 is necessary.

Parameters

count	Number of elements to allocate memory for.
location	Location on where to allocate the memory.
device	(Optional) device id for memory allocation on accelerators.

Definition at line 99 of file accelerator.h.

References tarch::internal::allocateRawData().

Referenced by toolbox::blockstructured::interpolateCellDataAssociatedToVolumesIntoOverlappingCell_secondOrder(), peano4::stacks::STDVectorStackOverSmartPointers< T >::prepareMetaDataBuffers(), toolbox::blockstructured::tests::IandRVolumetricTest::testMatrixInterpolation(), and exahype2::fv::musclhancock::timeStepWithMusclHancock_patchwise_heap_functors().

Here is the call graph for this function:

Here is the caller graph for this function:

enableNonCriticalAssertions()

void tarch::enableNonCriticalAssertions

(

bool

value

)

Switch noncritical assertions on/off.

Definition at line 39 of file NonCriticalAssertions.cpp.

freeMemory()

void tarch::freeMemory	(	void *	data,
		MemoryLocation	location,
		int	device = accelerator::Device::HostDevice )

Free memory.

This routine is thread-safe in the sense that it checks if data actually point to something that is not nullptr. If data is the nullptr, the operation degenerates to nop. This is important as we use the routine within smart pointers, where we don't want to check manually if data is actually nullptr or not.

Definition at line 30 of file accelerator.cpp.

Referenced by exahype2::fd::applySommerfeldConditions(), exahype2::EnclaveBookkeeping::cancelTask(), toolbox::blockstructured::tests::IandRInMatricesTest::compareInterpolateLinearCaseToTensorProductPatch(), toolbox::blockstructured::tests::IandRInMatricesTest::compareRestrictLinearCaseToTensorProductPatch(), tarch::la::DynamicMatrix::insertEmptyColumns(), tarch::la::DynamicMatrix::insertEmptyRows(), toolbox::blockstructured::interpolateHaloLayer_AoS_linear_with_constant_extrapolation(), toolbox::blockstructured::interpolateHaloLayer_AoS_linear_with_constant_extrapolation_and_linear_normal_interpolation(), toolbox::blockstructured::interpolateHaloLayer_AoS_linear_with_linear_extrapolation(), toolbox::blockstructured::interpolateHaloLayer_AoS_linear_with_linear_extrapolation_and_linear_normal_interpolation(), toolbox::blockstructured::interpolateHaloLayer_AoS_piecewise_constant(), tarch::la::DynamicMatrix::removeColumn(), tarch::la::DynamicMatrix::replicateCols(), tarch::la::DynamicMatrix::replicateRows(), runBenchmark(), runBenchmarks(), toolbox::particles::memorypool::GlobalContinuousMemoryPool< T >::scatter(), toolbox::particles::memorypool::VertexWiseContinuousMemoryPool< T >::scatter(), toolbox::particles::memorypool::GlobalContinuousMemoryPool< T >::scatterAndUpdateIterator(), toolbox::particles::memorypool::VertexWiseContinuousMemoryPool< T >::scatterAndUpdateIterator(), tarch::la::DynamicMatrix::shiftColumnsRight(), tarch::la::DynamicMatrix::shiftRowsDown(), exahype2::fv::rusanov::tests::ApplySplit1DRiemannToPatchTest::testIterateGrid(), toolbox::blockstructured::tests::IandRVolumetricTest::testMatrixInterpolation(), exahype2::fd::fd4::timeStep_patchwise_heap_functors(), exahype2::fv::musclhancock::timeStepWithMusclHancock_patchwise_heap_functors(), peano4::stacks::STDVectorStackOverSmartPointers< T >::tryToFinishSendOrReceive(), exahype2::EnclaveBookkeeping::waitForTaskAndDiscardResult(), exahype2::EnclaveBookkeeping::waitForTaskToTerminateAndCopyResultOver(), exahype2::CellData< inType, outType >::~CellData(), and tarch::la::DynamicMatrix::~DynamicMatrix().

Here is the caller graph for this function:

getFreeMemory()

int tarch::getFreeMemory

(

MemoryUsageFormat

format

)

Definition at line 83 of file tarch.cpp.

References getTotalMemory(), and MByte.

Referenced by toolbox::loadbalancing::AbstractLoadBalancing::fitsIntoMemory(), and toolbox::loadbalancing::strategies::SpreadOutHierarchically::getNumberOfSplitsOnLocalRank().

Here is the call graph for this function:

Here is the caller graph for this function:

getMemoryUsage()

int tarch::getMemoryUsage

(

MemoryUsageFormat

format

)

Method for getting the application's memory footprint.

We use the exact version of KiloByte with 1024 bytes.

The implementation is based on C-version of Michael Moltenbrey.

Definition at line 95 of file tarch.cpp.

References _log, f(), and logError.

Referenced by toolbox::loadbalancing::AbstractLoadBalancing::fitsIntoMemory(), toolbox::loadbalancing::strategies::SpreadOutHierarchically::getNumberOfSplitsOnLocalRank(), and peano4::parallel::SpacetreeSet::split().

Here is the call graph for this function:

Here is the caller graph for this function:

getTotalMemory()

int tarch::getTotalMemory

(

MemoryUsageFormat

format

)

Definition at line 72 of file tarch.cpp.

Referenced by getFreeMemory(), and peano4::parallel::SpacetreeSet::split().

Here is the caller graph for this function:

getUnitTests()

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

(

)

Please destroy after usage.

Definition at line 17 of file UnitTests.cpp.

References tarch::tests::TreeTestCaseCollection::addTestCase().

Referenced by main(), and runTests().

Here is the call graph for this function:

Here is the caller graph for this function:

gpuAbort()

void tarch::gpuAbort

(

)

Delegates to std::abort() if no GPU offloading is active.

Otherwise is degenerates to nop, as there is no abort on an accelerator. The vanilla implementation can be found in accelerator.cpp. The vendor-specific implementations are hold within their accelerator.cpp file.

Definition at line 36 of file accelerator.cpp.

hasNonCriticalAssertionBeenViolated()

bool tarch::hasNonCriticalAssertionBeenViolated

(

)

Definition at line 91 of file NonCriticalAssertions.cpp.

References _log, tarch::mpi::Rank::getCommunicator(), tarch::mpi::Rank::getInstance(), tarch::mpi::IntegerMessage::getValue(), logError, and tarch::mpi::IntegerMessage::receive().

Referenced by selectNextAlgorithmicStep().

Here is the call graph for this function:

Here is the caller graph for this function:

initNonCriticalAssertionEnvironment()

void tarch::initNonCriticalAssertionEnvironment

(

)

Register the assertion tag from the global communicator.

Definition at line 51 of file NonCriticalAssertions.cpp.

References tarch::mpi::Rank::reserveFreeTag().

Referenced by main().

Here is the call graph for this function:

Here is the caller graph for this function:

memset()

double * tarch::memset	(	double *	dest,
		double	ch,
		size_t	byteCount )

Alternative GPU-ready version of memset.

AMD's standard library hasn't annotated the memset call with target offload for OpenMP. So we have to provide a GPU-ready version ourselves. We do so if and only if the OpenMPManuallyOffloadMemset macro is set. It is typically set in the CompilerSpecificSettings.h. If that macro is not defined, we fall back to std::memset.

The signature here is compatible with std::memset. The original interface is

char* memset(char* dest, int ch, size_t count);

which is just not nice. It should be

double* memset(double* dest, double ch, size_t countNumberOfDoubles);

but that's not what the C++ standard gives us. So we pick the nicest interface we could think of, but is still compatible.

Parameters

count

Byte count.

Definition at line 54 of file accelerator.cpp.

References assertion2, and double.

Referenced by applications::exahype2::ccz4::admconstraints(), exahype2::fd::tests::CCZ4KernelTest::AppleWithAppleTest(), applications::exahype2::ccz4::diagonal_gaugeWave(), applications::exahype2::ccz4::flat(), applications::exahype2::ccz4::gaugeWave(), benchmarks::exahype2::ccz4::CCZ4::initialCondition(), applications::exahype2::ccz4::linearWave(), applications::exahype2::ccz4::CCZ4::sourceTerm(), and applications::exahype2::ccz4::TestingOutput().

Here is the caller graph for this function:

padSizeToAlignment()

std::size_t tarch::padSizeToAlignment	(	std::size_t	size,
		std::size_t	alignment )

Definition at line 40 of file accelerator.cpp.

References assertion.

Referenced by tarch::internal::allocateRawData().

Here is the caller graph for this function:

shutdownNonCriticalAssertionEnvironment()

void tarch::shutdownNonCriticalAssertionEnvironment

(

)

peano4::shutdownParallelEnvironment().

Definition at line 44 of file NonCriticalAssertions.cpp.

References tarch::mpi::Rank::releaseTag().

Referenced by main().

Here is the call graph for this function:

Here is the caller graph for this function:

toString()

std::string tarch::toString

(

MemoryLocation

value

)

Definition at line 7 of file accelerator.cpp.

Referenced by tarch::la::DynamicMatrix::insertEmptyColumns(), tarch::la::DynamicMatrix::insertEmptyRows(), tarch::timing::Measurement::max(), tarch::timing::Measurement::min(), tarch::la::Vector< Size, Scalar >::operator()(), tarch::la::Vector< Size, Scalar >::operator()(), tarch::la::Vector< Size, Scalar >::operator[](), tarch::la::Vector< Size, Scalar >::operator[](), tarch::la::tests::VectorTest::testAssignment(), tarch::la::tests::VectorTest::testVectorScalarOperations(), and tarch::la::tests::VectorTest::testWrappedVector().

Here is the caller graph for this function:

triggerNonCriticalAssertion()

void tarch::triggerNonCriticalAssertion	(	std::string	file,
		int	line,
		std::string	expression,
		std::string	parameterValuePairs )

Trigger a non-critical assertion.

You can either use this routine, or you can use the macros. They invoke this routine eventually.

The routine reports only details about the very first non-critical assertion. All follow-up logs on this rank are ignored, i.e. we only get a note and no detail. Once logged locally, we send the assertion message to the global master if we are not working on the global marker ourselves.

Definition at line 58 of file NonCriticalAssertions.cpp.

References _log, tarch::mpi::Rank::getGlobalMasterRank(), tarch::mpi::Rank::getInstance(), tarch::mpi::Rank::getRank(), logDebug, logError, and tarch::mpi::IntegerMessage::send().

Referenced by applications::exahype2::euler::sphericalaccretion::SSInfall::initialCondition().

Here is the call graph for this function:

Here is the caller graph for this function: