TNL::Containers::DistributedVector< Real, Device, Index, Allocator > Class Template Reference

DistributedVector extends DistributedArray with algebraic operations. More...

#include <TNL/Containers/DistributedVector.h>

Inheritance diagram for TNL::Containers::DistributedVector< Real, Device, Index, Allocator >:

Collaboration diagram for TNL::Containers::DistributedVector< Real, Device, Index, Allocator >:

Public Types
using	AllocatorType = Allocator
using	ConstLocalViewType = Containers::VectorView< std::add_const_t< Real >, Device, Index >
using	ConstViewType = DistributedVectorView< std::add_const_t< Real >, Device, Index >
using	DeviceType = Device
using	IndexType = Index
using	LocalViewType = Containers::VectorView< Real, Device, Index >
using	RealType = Real
template<typename _Real, typename _Device = Device, typename _Index = Index, typename _Allocator = typename Allocators::Default< _Device >::template Allocator< _Real >>
using	Self = DistributedVector< _Real, _Device, _Index, _Allocator >
	A template which allows to quickly obtain a Vector type with changed template parameters.
using	ViewType = DistributedVectorView< Real, Device, Index >
Public Types inherited from TNL::Containers::DistributedArray< Real, Devices::Host, int, typename Allocators::Default< Devices::Host >::template Allocator< Real > >
using	AllocatorType
using	ConstLocalViewType
using	ConstViewType
using	DeviceType
using	IndexType
using	LocalRangeType
using	LocalViewType
using	Self
	A template which allows to quickly obtain a DistributedArray type with changed template parameters.
using	SynchronizerType
using	ValueType
using	ViewType

Public Member Functions
	DistributedVector (const DistributedVector &)=default
	Copy constructor (makes a deep copy).
	DistributedVector (const DistributedVector &vector, const AllocatorType &allocator)
	Copy constructor with a specific allocator (makes a deep copy).
	DistributedVector (DistributedVector &&) noexcept=default
	Default move constructor.
	DistributedArray ()=default
	Constructs an empty array with zero size.
	DistributedArray (const AllocatorType &allocator)
	Constructs an empty array and sets the provided allocator.
	DistributedArray (const DistributedArray &array)
	Copy constructor (makes a deep copy).
	DistributedArray (const DistributedArray &array, const AllocatorType &allocator)
	Copy constructor with a specific allocator (makes a deep copy).
	DistributedArray (DistributedArray &&) noexcept=default
	DistributedArray (LocalRangeType localRange, Index ghosts, Index globalSize, const MPI::Comm &communicator, const AllocatorType &allocator=AllocatorType())
ConstLocalViewType	getConstLocalView () const
	Returns a non-modifiable view of the local part of the vector.
ConstLocalViewType	getConstLocalViewWithGhosts () const
	Returns a non-modifiable view of the local part of the vector, including ghost values.
ConstViewType	getConstView () const
	Returns a non-modifiable view of the vector.
LocalViewType	getLocalView ()
	Returns a modifiable view of the local part of the vector.
LocalViewType	getLocalViewWithGhosts ()
	Returns a modifiable view of the local part of the vector, including ghost values.
ViewType	getView ()
	Returns a modifiable view of the vector.
	operator ConstViewType () const
	Conversion operator to a non-modifiable view of the vector.
	operator ViewType ()
	Conversion operator to a modifiable view of the vector.
template<typename Vector, typename..., typename = std::enable_if_t< HasSubscriptOperator< Vector >::value >>
DistributedVector &	operator%= (const Vector &vector)
template<typename Vector, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator%= (const Vector &vector)
template<typename Scalar, typename..., typename = std::enable_if_t< ! HasSubscriptOperator< Scalar >::value >>
DistributedVector &	operator%= (Scalar c)
template<typename Scalar, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator%= (Scalar c)
template<typename Vector, typename..., typename = std::enable_if_t< HasSubscriptOperator< Vector >::value >>
DistributedVector &	operator*= (const Vector &vector)
template<typename Vector, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator*= (const Vector &vector)
template<typename Scalar, typename..., typename = std::enable_if_t< ! HasSubscriptOperator< Scalar >::value >>
DistributedVector &	operator*= (Scalar c)
template<typename Scalar, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator*= (Scalar c)
template<typename Vector, typename..., typename = std::enable_if_t< HasSubscriptOperator< Vector >::value >>
DistributedVector &	operator+= (const Vector &vector)
template<typename Vector, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator+= (const Vector &vector)
template<typename Scalar, typename..., typename = std::enable_if_t< ! HasSubscriptOperator< Scalar >::value >>
DistributedVector &	operator+= (Scalar c)
template<typename Scalar, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator+= (Scalar c)
template<typename Vector, typename..., typename = std::enable_if_t< HasSubscriptOperator< Vector >::value >>
DistributedVector &	operator-= (const Vector &vector)
template<typename Vector, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator-= (const Vector &vector)
template<typename Scalar, typename..., typename = std::enable_if_t< ! HasSubscriptOperator< Scalar >::value >>
DistributedVector &	operator-= (Scalar c)
template<typename Scalar, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator-= (Scalar c)
template<typename Vector, typename..., typename = std::enable_if_t< HasSubscriptOperator< Vector >::value >>
DistributedVector &	operator/= (const Vector &vector)
template<typename Vector, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator/= (const Vector &vector)
template<typename Scalar, typename..., typename = std::enable_if_t< ! HasSubscriptOperator< Scalar >::value >>
DistributedVector &	operator/= (Scalar c)
template<typename Scalar, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator/= (Scalar c)
DistributedVector &	operator= (const DistributedVector &)=default
	Copy-assignment operator for copying data from another vector.
template<typename Vector, typename..., typename = std::enable_if_t< HasSubscriptOperator< Vector >::value >>
DistributedVector &	operator= (const Vector &vector)
template<typename Vector, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator= (const Vector &vector)
DistributedVector &	operator= (DistributedVector &&) noexcept(false)=default
	Move-assignment operator for acquiring data from rvalues.
template<typename Scalar, typename..., typename = std::enable_if_t< ! HasSubscriptOperator< Scalar >::value >>
DistributedVector &	operator= (Scalar c)
template<typename Scalar, typename..., typename>
DistributedVector< Real, Device, Index, Allocator > &	operator= (Scalar c)
Public Member Functions inherited from TNL::Containers::DistributedArray< Real, Devices::Host, int, typename Allocators::Default< Devices::Host >::template Allocator< Real > >
	DistributedArray ()=default
	Constructs an empty array with zero size.
void	copyFromGlobal (ConstLocalViewType globalArray)
bool	empty () const
void	forElements (IndexType begin, IndexType end, Function &&f)
	Process the lambda function f for each array element in interval [ begin, end).
AllocatorType	getAllocator () const
	Returns the allocator associated to the array.
const MPI::Comm &	getCommunicator () const
	Returns the MPI communicator associated to the array.
ConstLocalViewType	getConstLocalView () const
	Returns a non-modifiable view of the local part of the array.
ConstLocalViewType	getConstLocalViewWithGhosts () const
	Returns a non-modifiable view of the local part of the array, including ghost values.
ConstViewType	getConstView () const
	Returns a non-modifiable view of the array.
ValueType	getElement (IndexType i) const
IndexType	getGhosts () const
const LocalRangeType &	getLocalRange () const
	Returns the local range of the distributed array.
LocalViewType	getLocalView ()
	Returns a modifiable view of the local part of the array.
LocalViewType	getLocalViewWithGhosts ()
	Returns a modifiable view of the local part of the array, including ghost values.
IndexType	getSize () const
std::shared_ptr< SynchronizerType >	getSynchronizer () const
int	getValuesPerElement () const
ViewType	getView ()
	Returns a modifiable view of the array.
void	loadFromGlobalFile (const String &fileName, bool allowCasting=false)
	operator ConstViewType () const
	Conversion operator to a non-modifiable view of the array.
	operator ViewType ()
	Conversion operator to a modifiable view of the array.
bool	operator!= (const Array &array) const
DistributedArray< Real, Devices::Host, int, Allocator > &	operator= (const Array &array)
bool	operator== (const Array &array) const
__cuda_callable__ ValueType &	operator[] (IndexType i)
void	reset ()
void	setDistribution (LocalRangeType localRange, int ghosts, int globalSize, const MPI::Comm &communicator)
	Set new global size and distribution of the array.
void	setElement (IndexType i, ValueType value)
void	setLike (const Array &array)
void	setSynchronizer (std::shared_ptr< SynchronizerType > synchronizer, int valuesPerElement=1)
void	setValue (ValueType value)
void	startSynchronization ()
void	waitForSynchronization () const

Additional Inherited Members
Protected Attributes inherited from TNL::Containers::DistributedArray< Real, Devices::Host, int, typename Allocators::Default< Devices::Host >::template Allocator< Real > >
LocalArrayType	localData
ViewType	view

Detailed Description

template<typename Real, typename Device = Devices::Host, typename Index = int, typename Allocator = typename Allocators::Default< Device >::template Allocator< Real >>
class TNL::Containers::DistributedVector< Real, Device, Index, Allocator >

DistributedVector extends DistributedArray with algebraic operations.

Example

#include <iostream>
#include <TNL/Containers/BlockPartitioning.h>
#include <TNL/Containers/DistributedVector.h>
#include <TNL/MPI/ScopedInitializer.h>
 
/***
 * The following works for any device (CPU, GPU ...).
 */
template< typename Device >
void
distributedVectorExample()
{
   using VectorType = TNL::Containers::DistributedVector< int, Device >;
   using IndexType = typename VectorType::IndexType;
   using LocalRangeType = typename VectorType::LocalRangeType;
 
   const TNL::MPI::Comm communicator = MPI_COMM_WORLD;
 
   // We set the global vector size to a prime number to force non-uniform distribution.
   const int size = 97;
   const int ghosts = ( communicator.size() > 1 ) ? 4 : 0;
 
   const LocalRangeType localRange = TNL::Containers::splitRange< IndexType >( size, communicator );
   VectorType v( localRange, ghosts, size, communicator );
   v.forElements( 0,
                  size,
                  [] __cuda_callable__( int idx, int& value )
                  {
                     value = idx;
                  } );
   std::cout << "Rank " << communicator.rank() << " has subrange " << localRange << std::endl;
   const int sum = TNL::sum( v );
 
   if( communicator.rank() == 0 )
      std::cout << "Global sum is " << sum << std::endl;
}
 
int
main( int argc, char* argv[] )
{
   TNL::MPI::ScopedInitializer mpi( argc, argv );
 
   if( TNL::MPI::GetRank() == 0 )
      std::cout << "The first test runs on CPU ..." << std::endl;
   distributedVectorExample< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   TNL::MPI::Barrier();
 
   if( TNL::MPI::GetRank() == 0 )
      std::cout << "The second test runs on GPU ..." << std::endl;
   distributedVectorExample< TNL::Devices::Cuda >();
#endif
}

Output

Rank 2: rank on node is 2, using GPU id 0 of 2
Environment:
  CUDA_VISIBLE_DEVICES=
Rank 0: rank on node is 0, using GPU id 0 of 2
Environment:
  CUDA_VISIBLE_DEVICES=
Rank 1: rank on node is 1, using GPU id 1 of 2
Environment:
  CUDA_VISIBLE_DEVICES=
Rank 3: rank on node is 3, using GPU id 1 of 2
Environment:
  CUDA_VISIBLE_DEVICES=
Rank 1 has subrange TNL::Containers::Subrange<int>( 25, 49 )
Rank 3 has subrange TNL::Containers::Subrange<int>( 73, 97 )
Rank 2 has subrange TNL::Containers::Subrange<int>( 49, 73 )
The first test runs on CPU ...
Rank 0 has subrange TNL::Containers::Subrange<int>( 0, 25 )
Global sum is 4656
The second test runs on GPU ...
Rank 1 has subrange TNL::Containers::Subrange<int>( 25, 49 )
Rank 0 has subrange TNL::Containers::Subrange<int>( 0, 25 )
Rank 3 has subrange TNL::Containers::Subrange<int>( 73, 97 )
Rank 2 has subrange TNL::Containers::Subrange<int>( 49, 73 )
Global sum is 4656

Member Function Documentation

DistributedArray() [1/3]

template<typename Real, typename Device = Devices::Host, typename Index = int, typename Allocator = typename Allocators::Default< Device >::template Allocator< Real >>

TNL::Containers::DistributedArray< Real, Device, Index, Allocator >::DistributedArray ( const AllocatorType & allocator )

explicit

Constructs an empty array and sets the provided allocator.

Parameters

allocator

The allocator to be associated with this array.

DistributedArray() [2/3]

template<typename Real, typename Device = Devices::Host, typename Index = int, typename Allocator = typename Allocators::Default< Device >::template Allocator< Real >>

TNL::Containers::DistributedArray< Real, Device, Index, Allocator >::DistributedArray ( const DistributedArray & array )

explicit

Copy constructor (makes a deep copy).

Parameters

array

The array to be copied.

DistributedArray() [3/3]

template<typename Real, typename Device = Devices::Host, typename Index = int, typename Allocator = typename Allocators::Default< Device >::template Allocator< Real >>

TNL::Containers::DistributedArray< Real, Device, Index, Allocator >::DistributedArray ( const DistributedArray & array, const AllocatorType & allocator )

explicit

Copy constructor with a specific allocator (makes a deep copy).

Parameters

array	The array to be copied.
allocator	The allocator to be associated with this array.

---

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

• src/TNL/Containers/DistributedVector.h
• src/TNL/Containers/DistributedVector.hpp