lama::LAMAArray< T > Class Template Reference

LAMAArray is the base array containter for all compute relevant data within LAMA. More...

#include <LAMAArray.hpp>

Inheritance diagram for lama::LAMAArray< T >:

Public Types
typedef T	ValueType
	ValueType is the type stored in this container.
Public Member Functions
	LAMAArray ()
	LAMAArray() creates an empty LAMAArray with size 0.
	LAMAArray (const IndexType n)
	LAMAArray( const IndexType n ) creates a LAMAArray of size n.
	LAMAArray (const IndexType n, const ValueType &value)
	creates a LAMAArray of size n
template<typename OtherValueType >
	LAMAArray (const IndexType n, const OtherValueType *const values)
	creates a LAMAArray of size n
	LAMAArray (const LAMAArray< ValueType > &other)
	creates a copy of the passed LAMAArray
virtual	~LAMAArray ()
	Releases all used resources.
LAMAArray< ValueType > &	operator= (const LAMAArray< ValueType > &other)
	Copies the passed LAMAArray into this.
void	assign (const LAMAArray< ValueType > &other, ContextPtr context)
	Copies the passed LAMAArray into this.
void	swap (LAMAArray< ValueType > &other)
	swaps the contens of this with other
void	prefetch (ContextPtr context) const
	prefetches the contents of the container to the passed context
bool	isAvailableAt (ContextPtr context) const
	Checks if the data of this LAMAArray is available at the passed context.
ContextPtr	getValidContext (const Context::ContextType preferredType=Context::Host) const
	Gets the fist context where the data of this LAMAArray is available.
void	wait () const
	waits for potentially running prefetch
void	clear ()
	sets the size of this to 0 but does not free any memory
void	purge ()
	sets the size of this to 0 an frees all memory
virtual void	writeAt (std::ostream &stream) const
	writes the current status of this to the passed stream
virtual Scalar::ScalarType	getValueType () const
	Implementation of pure method.
IndexType	size () const
	Query the current size of the LAMA array, i.e.
Static Public Member Functions
static std::auto_ptr< _LAMAArray >	create (const Scalar::ScalarType)
Protected Member Functions
ValueType *	get (size_t index)
const ValueType *	get (const size_t index) const
void	clear (const size_t index)
void	resize (const size_t index, const IndexType newSize)
void	reserve (const size_t index, const IndexType capacity, const bool copy) const
IndexType	capacity (const size_t index) const
int	acquireReadAccess (ContextPtr context) const
	Complete handling to get read access for a certain context.
void	releaseReadAccess (const size_t index) const
int	acquireWriteAccess (ContextPtr context, bool keepFlag)
	Complete handling to get write access for a certain context.
int	acquireWriteAccess ()
void	releaseWriteAccess (const size_t index)
void	copy (const int toIndex, const int fromIndex) const
void	fetch (Context::ContextData &target, const Context::ContextData &source) const
	Copy data from a valid source context to an invalid target context.
std::auto_ptr< SyncToken >	fetchAsync (Context::ContextData &target, const Context::ContextData &source) const
	Asynchronous copy data from a valid source context to an invalid target context.
void	getAccess (size_t &contextIndex, size_t &validIndex, ContextPtr context, Context::ContextData::AccessKind kind) const
void	setHostContext ()
	LAMA_LOG_DECL_STATIC_LOGGER (logger)
Protected Attributes
std::vector < Context::ContextData * >	mContextData
std::auto_ptr< SyncToken >	mSyncToken
	outstanding transfers
boost::recursive_mutex	access_mutex
IndexType	mSize
	number of entries for the context array, common for all contexts
bool	constFlag
	if true the array cannot be written
Static Protected Attributes
static size_t	nContextIndex = std::numeric_limits<std::size_t>::max()
Friends
class	LAMAArrayView< T >
class	LAMAArrayConstView< T >

---

Detailed Description

template<typename T>
class lama::LAMAArray< T >

LAMAArray is the base array containter for all compute relevant data within LAMA.

LAMAArray its contens on all supported Locations, e.g. Host, CUDA and OpenCL. It transparently handles synchronization between the Locations. To enforce the consitency of the data a LAMAArray can be only indirectly accessed via a ReadAccess or a WriteAccess.

---

Member Typedef Documentation

template<typename T>

typedef T lama::LAMAArray< T >::ValueType

ValueType is the type stored in this container.

---

Constructor & Destructor Documentation

template<typename ValueType >

lama::LAMAArray< ValueType >::LAMAArray

(

)

LAMAArray() creates an empty LAMAArray with size 0.

References lama::LAMAArray< T >::setHostContext().

template<typename ValueType >

lama::LAMAArray< ValueType >::LAMAArray

(

const IndexType

n

)

[explicit]

LAMAArray( const IndexType n ) creates a LAMAArray of size n.

Parameters:

[in]

n

the size of the LAMAArray to create

LAMAArray( const IndexType n ) creates a LAMAArray of size n and allocates uninitialized Host memory.

References lama::Context::ContextData::allocate(), lama::LAMAArray< T >::mContextData, lama::_LAMAArray::mSize, and lama::LAMAArray< T >::setHostContext().

template<typename ValueType >

lama::LAMAArray< ValueType >::LAMAArray	(	const IndexType	n,
		const ValueType &	value
	)

creates a LAMAArray of size n

Parameters:

[in]	n	the size of the LAMAArray to create
[in]	value	the value to initialize the container contens with

LAMAArray( const IndexType n ) creates a LAMAArray of size n, allocates Host memory and fills the Host memory with the passed value.

References lama::Context::ContextData::allocate(), lama::LAMAArray< T >::mContextData, lama::_LAMAArray::mSize, lama::Context::ContextData::pointer, lama::LAMAArray< T >::setHostContext(), and lama::Context::ContextData::valid.

template<typename ValueType >

template<typename OtherValueType >

lama::LAMAArray< ValueType >::LAMAArray	(	const IndexType	n,
		const OtherValueType *const	values
	)

creates a LAMAArray of size n

Parameters:

[in]	n	the size of the LAMAArray to create
[in]	values	the values to initialize the container contens with

LAMAArray( const IndexType n ) creates a LAMAArray of size n, allocates Host memory and fills the Host memory with the passed values.

References lama::Context::ContextData::allocate(), lama::LAMAArray< T >::mContextData, lama::_LAMAArray::mSize, lama::Context::ContextData::pointer, lama::LAMAArray< T >::setHostContext(), and lama::Context::ContextData::valid.

template<typename ValueType >

lama::LAMAArray< ValueType >::LAMAArray

(

const LAMAArray< ValueType > &

other

)

creates a copy of the passed LAMAArray

Parameters:

[in]

other

the LAMAArray to copy

LAMAArray(const LAMAArray<ValueType>& other) copies the passed LAMAArray. The container contens is copied for all currently valid Locations.

References lama::LAMAArray< T >::operator=(), and lama::LAMAArray< T >::setHostContext().

template<typename ValueType >

lama::LAMAArray< ValueType >::~LAMAArray

(

)

[virtual]

Releases all used resources.

---

Member Function Documentation

template<typename ValueType >

int lama::LAMAArray< ValueType >::acquireReadAccess

(

ContextPtr

context

)

const [protected]

Complete handling to get read access for a certain context.

Returns:

index of context data array that contains the valid entry.

References lama::Context::ContextData::context, LAMA_ASSERT_DEBUG, lama::Context::ContextData::lock, and lama::Context::ContextData::valid.

template<typename ValueType >

int lama::LAMAArray< ValueType >::acquireWriteAccess	(	ContextPtr	context,
		bool	keepFlag
	)		[protected]

Complete handling to get write access for a certain context.

Returns:

index of context data array that contains the valid entry.

References lama::Context::ContextData::context, LAMA_ASSERT_DEBUG, lama::Context::ContextData::lock, and lama::Context::ContextData::valid.

template<typename ValueType >

int lama::LAMAArray< ValueType >::acquireWriteAccess

(

)

[protected]

References lama::Context::ContextData::context, lama::max(), and lama::Context::ContextData::valid.

template<typename ValueType >

void lama::LAMAArray< ValueType >::assign	(	const LAMAArray< ValueType > &	other,
		ContextPtr	context
	)

Copies the passed LAMAArray into this.

Parameters:

[in]	other	the LAMAArray to copy
[in]	context	the context where the assignment should be carried out

The assign method copies the passed LAMAArray. The container content is copied for the passed contexts. If necessary other is copied to context to carry this out.

References lama::Context::ContextData::context, LAMA_ASSERT_ERROR, LAMA_THROWEXCEPTION, lama::Context::ContextData::lock, lama::LAMAArray< T >::mContextData, lama::_LAMAArray::mSize, lama::LAMAArray< T >::prefetch(), lama::Context::ContextData::valid, and lama::LAMAArray< T >::wait().

template<typename ValueType >

IndexType lama::LAMAArray< ValueType >::capacity

(

const size_t

index

)

const [protected]

References LAMA_ASSERT_DEBUG.

template<typename ValueType >

void lama::LAMAArray< ValueType >::clear

(

)

sets the size of this to 0 but does not free any memory

This operation should be used before a write-only access that is followed by a resize of the array. It avoids data transfer between two contextes.

References LAMA_ASSERT.

Referenced by lama::_MatrixStorage::_assign(), lama::_MatrixStorage::_assignTranspose(), lama::Halo::clear(), and lama::Communicator::shiftAsync().

template<typename ValueType >

void lama::LAMAArray< ValueType >::clear

(

const size_t

index

)

[protected]

References LAMA_ASSERT_DEBUG, and LAMA_ASSERT_ERROR.

template<typename ValueType >

void lama::LAMAArray< ValueType >::copy	(	const int	toIndex,
		const int	fromIndex
	)		const [protected]

References LAMA_THROWEXCEPTION.

std::auto_ptr< _LAMAArray > lama::_LAMAArray::create

(

const Scalar::ScalarType

valueType

)

[static, inherited]

References lama::Scalar::DOUBLE, lama::Scalar::FLOAT, and LAMA_THROWEXCEPTION.

template<typename ValueType >

void lama::LAMAArray< ValueType >::fetch	(	Context::ContextData &	target,
		const Context::ContextData &	source
	)		const [protected]

Copy data from a valid source context to an invalid target context.

This routine assumes that there is already sufficient memory allocated in the target context.

References lama::Context::ContextData::context, LAMA_ASSERT, LAMA_ASSERT_DEBUG, lama::Context::ContextData::pointer, lama::Context::ContextData::size, and lama::Context::ContextData::valid.

template<typename ValueType >

std::auto_ptr< SyncToken > lama::LAMAArray< ValueType >::fetchAsync	(	Context::ContextData &	target,
		const Context::ContextData &	source
	)		const [protected]

Asynchronous copy data from a valid source context to an invalid target context.

References lama::Context::ContextData::context, LAMA_ASSERT_DEBUG, LAMA_THROWEXCEPTION, lama::Context::ContextData::pointer, lama::Context::ContextData::size, and lama::Context::ContextData::valid.

template<typename ValueType >

ValueType * lama::LAMAArray< ValueType >::get

(

size_t

index

)

[protected]

References LAMA_ASSERT_DEBUG.

template<typename ValueType >

const ValueType * lama::LAMAArray< ValueType >::get

(

const size_t

index

)

const [protected]

References LAMA_ASSERT_DEBUG.

template<typename ValueType >

void lama::LAMAArray< ValueType >::getAccess	(	size_t &	contextIndex,
		size_t &	validIndex,
		ContextPtr	context,
		Context::ContextData::AccessKind	kind
	)		const [protected]

References lama::Context::ContextData::context, LAMA_ASSERT_DEBUG, LAMA_THROWEXCEPTION, lama::Context::ContextData::lock, and lama::Context::ContextData::valid.

template<typename ValueType >

ContextPtr lama::LAMAArray< ValueType >::getValidContext

(

const Context::ContextType

preferredType = Context::Host

)

const

Gets the fist context where the data of this LAMAArray is available.

If possible a context of the passed preferred type is returned.

Parameters:

[in]

preferredType

the preferred type for the valid context.

Returns:

a context there the data of this LAMAArray is available.

References lama::Context::ContextData::context, LAMA_ASSERT_DEBUG, lama::Context::ContextData::lock, and lama::Context::ContextData::valid.

Referenced by lama::LAMAArrayUtils::gather().

template<typename ValueType >

Scalar::ScalarType lama::LAMAArray< ValueType >::getValueType

(

)

const [virtual]

Implementation of pure method.

Implements lama::_LAMAArray.

Referenced by lama::LAMAArrayUtils::assignImpl1().

template<typename ValueType >

bool lama::LAMAArray< ValueType >::isAvailableAt

(

ContextPtr

context

)

const

Checks if the data of this LAMAArray is available at the passed context.

Parameters:

[in]

context

the context to check for availability of data.

Returns:

if the data of this LAMAArray is available at context

References LAMA_ASSERT_ERROR.

template<typename T>

lama::LAMAArray< T >::LAMA_LOG_DECL_STATIC_LOGGER

(

logger

)

[protected]

template<typename ValueType >

LAMAArray< ValueType > & lama::LAMAArray< ValueType >::operator=

(

const LAMAArray< ValueType > &

other

)

Copies the passed LAMAArray into this.

Parameters:

[in]

other

the LAMAArray to copy

The assignment operator copies the passed LAMAArray. The container content is copied for all contexts where a valid copy is available (at least one).

References lama::Context::ContextData::context, LAMA_ASSERT_ERROR, LAMA_THROWEXCEPTION, lama::Context::ContextData::lock, lama::LAMAArray< T >::mContextData, lama::_LAMAArray::mSize, lama::Context::ContextData::valid, and lama::LAMAArray< T >::wait().

Referenced by lama::LAMAArray< T >::LAMAArray().

template<typename ValueType >

void lama::LAMAArray< ValueType >::prefetch

(

ContextPtr

context

)

const

prefetches the contents of the container to the passed context

Parameters:

[in]

context

the context to prefetch to

This method prefetches the contents of the container to the context. If this valid at location nothing happens,if not a transfer from a valid location to the passed location is started. Because the transfer is handled by LAMAArray and to maintain the consistency of the container only one running transfer can exist at any point in time. There for if two prefetches to two different invalid locations are started one after the other the second transfer does not start before the first one is finished.

References LAMA_ASSERT_DEBUG, LAMA_ASSERT_ERROR, and lama::Context::ContextData::valid.

Referenced by lama::LAMAArray< T >::assign(), and lama::SpecializedJacobi::iterateTyped().

template<typename ValueType >

void lama::LAMAArray< ValueType >::purge

(

)

sets the size of this to 0 an frees all memory

References lama::Context::ContextData::free(), LAMA_ASSERT, and lama::Context::ContextData::lock.

template<typename ValueType >

void lama::LAMAArray< ValueType >::releaseReadAccess

(

const size_t

index

)

const [protected]

References LAMA_ASSERT, and LAMA_ASSERT_DEBUG.

template<typename ValueType >

void lama::LAMAArray< ValueType >::releaseWriteAccess

(

const size_t

index

)

[protected]

References LAMA_ASSERT, and LAMA_ASSERT_DEBUG.

template<typename ValueType >

void lama::LAMAArray< ValueType >::reserve	(	const size_t	index,
		const IndexType	capacity,
		const bool	copy
	)		const [protected]

References LAMA_ASSERT_DEBUG, and LAMA_ASSERT_ERROR.

template<typename ValueType >

void lama::LAMAArray< ValueType >::resize	(	const size_t	index,
		const IndexType	newSize
	)		[protected]

References LAMA_ASSERT_DEBUG, and LAMA_ASSERT_ERROR.

template<typename ValueType >

void lama::LAMAArray< ValueType >::setHostContext

(

)

[protected]

References LAMA_ASSERT_DEBUG, LAMA_MAX_CONTEXTS, and lama::Context::ContextData::valid.

Referenced by lama::LAMAArray< T >::LAMAArray().

IndexType lama::_LAMAArray::size

(

)

const [inline, inherited]

Query the current size of the LAMA array, i.e.

number of entries.

Returns:

the number of entries of the array.

References lama::_LAMAArray::mSize.

Referenced by lama::CommunicationPlan::allocate(), lama::LAMAArrayUtils::assign(), lama::DenseVector< T >::assign(), lama::MatrixStorage< T >::assign(), lama::LAMAArrayUtils::assignImpl2(), lama::MatrixStorage< T >::assignTranspose(), lama::SparseMatrix< T >::assignTransposeImpl(), lama::SparseAssemblyStorage< T >::buildCSR(), lama::_StorageMethods::buildHalo(), lama::MatrixStorage< T >::buildHalo(), lama::MatrixStorage< T >::convertCSR2CSC(), lama::COOStorage< T >::COOStorage(), lama::MatrixStorage< T >::copyTo(), lama::Redistributor::copyV(), lama::DenseStorage< T >::DenseStorage(), lama::DenseStorageView< T >::DenseStorageView(), lama::DenseVector< T >::DenseVector(), lama::Communicator::exchangeByPlan(), lama::Communicator::exchangeByPlanAsync(), lama::MatrixStorage< T >::exchangeHalo(), lama::StorageMethods< T >::exchangeHaloCSR(), lama::LAMAArrayUtils::gather(), lama::Redistributor::gatherV(), lama::_MatrixStorage::getMemoryUsage(), lama::DenseMatrix< T >::invertCyclic(), lama::SpecializedJacobi::iterateTyped(), lama::StorageMethods< T >::joinCSR(), lama::MatrixStorage< T >::joinHalo(), lama::LAMAArrayRef< T >::LAMAArrayRef(), lama::MatrixStorage< T >::localize(), lama::StorageMethods< T >::localizeCSR(), lama::DenseMatrix< T >::matrixTimesVectorImpl(), lama::SparseMatrix< T >::matrixTimesVectorImpl(), lama::_MatrixStorage::offsets2sizes(), lama::MatrixStorage< T >::readFromFile(), lama::Redistributor::redistribute(), lama::MatrixStorage< T >::redistribute(), lama::StorageMethods< T >::redistributeCSR(), lama::Redistributor::redistributeN(), lama::MatrixStorage< T >::replicate(), lama::StorageMethods< T >::replicateCSR(), lama::CRTPMatrixStorage< CSRStorage< T >, T >::scale(), lama::SparseAssemblyStorage< T >::scaleImpl(), lama::CSRStorage< T >::scaleImpl(), lama::Redistributor::scatterV(), lama::DenseMatrix< T >::setCSRDataLocal(), lama::CRTPMatrixStorage< CSRStorage< T >, T >::setDiagonal(), lama::CSRStorage< T >::setDiagonalImpl(), lama::DIAStorage< T >::setDiagonalImpl(), lama::SparseAssemblyStorage< T >::setDiagonalImpl(), lama::MatrixStorage< T >::setRawDenseData(), lama::DenseVector< T >::setValues(), lama::_MatrixStorage::sizes2offsets(), lama::StorageMethods< T >::splitCSR(), lama::MatrixStorage< T >::splitHalo(), lama::CSRStorage< T >::swap(), lama::StorageIO< T >::writeCSRToBinaryFile(), lama::StorageIO< T >::writeCSRToFile(), lama::StorageIO< T >::writeCSRToFormattedFile(), lama::StorageIO< T >::writeCSRToMMFile(), and lama::StorageIO< T >::writeCSRToXDRFile().

template<typename ValueType >

void lama::LAMAArray< ValueType >::swap

(

LAMAArray< ValueType > &

other

)

swaps the contens of this with other

Parameters:

[in]

other

the LAMAArray to swap the contens with.

References LAMA_ASSERT_ERROR, lama::LAMAArray< T >::mContextData, lama::_LAMAArray::mSize, and lama::LAMAArray< T >::mSyncToken.

Referenced by lama::Communicator::computeOwners(), lama::DenseVector< T >::redistribute(), and lama::_MatrixStorage::swap().

template<typename ValueType >

void lama::LAMAArray< ValueType >::wait

(

)

const

waits for potentially running prefetch

Referenced by lama::LAMAArray< T >::assign(), and lama::LAMAArray< T >::operator=().

template<typename ValueType >

void lama::LAMAArray< ValueType >::writeAt

(

std::ostream &

stream

)

const [virtual]

writes the current status of this to the passed stream

Parameters:

[out]

stream

the stream to write to

Reimplemented from Printable.

References lama::Context::ContextData::context, lama::Context::ContextData::lock, lama::Context::ContextData::size, and lama::Context::ContextData::valid.

---

Friends And Related Function Documentation

template<typename T>

friend class LAMAArrayConstView< T > [friend]

template<typename T>

friend class LAMAArrayView< T > [friend]

---

Field Documentation

template<typename T>

boost::recursive_mutex lama::LAMAArray< T >::access_mutex [mutable, protected]

bool lama::_LAMAArray::constFlag [protected, inherited]

if true the array cannot be written

Referenced by lama::LAMAArrayRef< T >::LAMAArrayRef(), and lama::LAMAArrayView< T >::LAMAArrayView().

template<typename T>

std::vector<Context::ContextData*> lama::LAMAArray< T >::mContextData [mutable, protected]

Referenced by lama::LAMAArray< T >::assign(), lama::LAMAArray< T >::LAMAArray(), lama::LAMAArrayRef< T >::LAMAArrayRef(), lama::LAMAArray< T >::operator=(), and lama::LAMAArray< T >::swap().

IndexType lama::_LAMAArray::mSize [protected, inherited]

number of entries for the context array, common for all contexts

Referenced by lama::LAMAArray< T >::assign(), lama::LAMAArray< T >::LAMAArray(), lama::LAMAArrayRef< T >::LAMAArrayRef(), lama::LAMAArray< T >::operator=(), lama::_LAMAArray::size(), and lama::LAMAArray< T >::swap().

template<typename T>

std::auto_ptr<SyncToken> lama::LAMAArray< T >::mSyncToken [mutable, protected]

outstanding transfers

Referenced by lama::LAMAArray< T >::swap().

template<typename T>

size_t lama::LAMAArray< ValueType >::nContextIndex = std::numeric_limits<std::size_t>::max() [static, protected]

Reimplemented from lama::_LAMAArray.

---

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

• /home/brandes/workspace/LAMA/src/lama/LAMAArray.hpp
• /home/brandes/workspace/LAMA/src/lama/LAMAArray.cpp