Added for[,All]Rows methods to DenseMatrix[,View]. (f69fc84c) · Commits · TNL / tnl-dev

Documentation/Tutorials/Matrices/tutorial_Matrices.md

+2 −0

Original line number	Diff line number	Diff line
		@@ -22,6 +22,8 @@
		1. [Matrix reader and writer](#matrix-reader-and-writer)
		8. [Appendix](#appendix)

		TODO: Add description of forRows and sequentialForRows.

		## Introduction

		TNL offers several types of matrices like dense (\ref TNL::Matrices::DenseMatrix), sparse (\ref TNL::Matrices::SparseMatrix), tridiagonal (\ref TNL::Matrices::TridiagonalMatrix), multidiagonal (\ref TNL::Matrices::MultidiagonalMatrix) and lambda matrices (\ref TNL::Matrices::LambdaMatrix). The sparse matrices can be symmetric to lower the memory requirements. The interfaces of given matrix types are designed to be as unified as possible to ensure that the user can easily switch between different matrix types while making no or only a little changes in the source code. All matrix types allows traversing all matrix elements and manipulate them using lambda functions as well as performing flexible reduction in matrix rows. The following text describes particular matrix types and their unified interface in details.

src/TNL/Matrices/DenseMatrix.h

+100 −0

Original line number	Diff line number	Diff line
		@@ -612,6 +612,106 @@ class DenseMatrix : public Matrix< Real, Device, Index, RealAllocator >
		template< typename Function >
		void forAllElements( Function& function );

		/**
		* \brief Method for parallel iteration over matrix rows from interval [ \e begin, \e end).
		*
		* In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method
		* \ref DenseMatrix::forElements where more than one thread can be mapped to each row.
		*
		* \tparam Function is type of the lambda function.
		*
		* \param begin defines beginning of the range [ \e begin,\e end ) of rows to be processed.
		* \param end defines ending of the range [ \e begin, \e end ) of rows to be processed.
		* \param function is an instance of the lambda function to be called for each row.
		*
		* ```
		* auto function = [] __cuda_callable__ ( RowViewType& row ) mutable { ... };
		* ```
		*
		* \e RowViewType represents matrix row - see \ref TNL::Matrices::DenseMatrix::RowViewType.
		*
		* \par Example
		* \include Matrices/DenseMatrix/DenseMatrixExample_forRows.cpp
		* \par Output
		* \include DenseMatrixExample_forRows.out
		*/
		template< typename Function >
		void forRows( IndexType begin, IndexType end, Function&& function );

		/**
		* \brief Method for parallel iteration over matrix rows from interval [ \e begin, \e end) for constant instances.
		*
		* In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method
		* \ref DenseMatrix::forElements where more than one thread can be mapped to each row.
		*
		* \tparam Function is type of the lambda function.
		*
		* \param begin defines beginning of the range [ \e begin,\e end ) of rows to be processed.
		* \param end defines ending of the range [ \e begin, \e end ) of rows to be processed.
		* \param function is an instance of the lambda function to be called for each row.
		*
		* ```
		* auto function = [] __cuda_callable__ ( RowViewType& row ) { ... };
		* ```
		*
		* \e RowViewType represents matrix row - see \ref TNL::Matrices::DenseMatrix::RowViewType.
		*
		* \par Example
		* \include Matrices/DenseMatrix/DenseMatrixExample_forRows.cpp
		* \par Output
		* \include DenseMatrixExample_forRows.out
		*/
		template< typename Function >
		void forRows( IndexType begin, IndexType end, Function&& function ) const;

		/**
		* \brief Method for parallel iteration over all matrix rows.
		*
		* In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method
		* \ref DenseMatrix::forAllElements where more than one thread can be mapped to each row.
		*
		* \tparam Function is type of the lambda function.
		*
		* \param function is an instance of the lambda function to be called for each row.
		*
		* ```
		* auto function = [] __cuda_callable__ ( RowViewType& row ) mutable { ... };
		* ```
		*
		* \e RowViewType represents matrix row - see \ref TNL::Matrices::DenseMatrix::RowViewType.
		*
		* \par Example
		* \include Matrices/DenseMatrix/DenseMatrixExample_forRows.cpp
		* \par Output
		* \include DenseMatrixExample_forRows.out
		*/
		template< typename Function >
		void forAllRows( Function&& function );

		/**
		* \brief Method for parallel iteration over all matrix rows for constant instances.
		*
		* In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method
		* \ref DenseMatrix::forAllElements where more than one thread can be mapped to each row.
		*
		* \tparam Function is type of the lambda function.
		*
		* \param function is an instance of the lambda function to be called for each row.
		*
		* ```
		* auto function = [] __cuda_callable__ ( RowViewType& row ) { ... };
		* ```
		*
		* \e RowViewType represents matrix row - see \ref TNL::Matrices::DenseMatrix::RowViewType.
		*
		* \par Example
		* \include Matrices/DenseMatrix/DenseMatrixExample_forRows.cpp
		* \par Output
		* \include DenseMatrixExample_forRows.out
		*/
		template< typename Function >
		void forAllRows( Function&& function ) const;

		/**
		* \brief Method for sequential iteration over all matrix rows for constant instances.
		*

src/TNL/Matrices/DenseMatrix.hpp

+52 −0

Original line number	Diff line number	Diff line
		@@ -449,6 +449,58 @@ forAllElements( Function& function )
		this->forElements( 0, this->getRows(), function );
		}

		template< typename Real,
		typename Device,
		typename Index,
		ElementsOrganization Organization,
		typename RealAllocator >
		template< typename Function >
		void
		DenseMatrix< Real, Device, Index, Organization, RealAllocator >::
		forRows( IndexType begin, IndexType end, Function&& function )
		{
		this->getView().forRows( begin, end, function );
		}

		template< typename Real,
		typename Device,
		typename Index,
		ElementsOrganization Organization,
		typename RealAllocator >
		template< typename Function >
		void
		DenseMatrix< Real, Device, Index, Organization, RealAllocator >::
		forRows( IndexType begin, IndexType end, Function&& function ) const
		{
		this->getConstView().forRows( begin, end, function );
		}

		template< typename Real,
		typename Device,
		typename Index,
		ElementsOrganization Organization,
		typename RealAllocator >
		template< typename Function >
		void
		DenseMatrix< Real, Device, Index, Organization, RealAllocator >::
		forAllRows( Function&& function )
		{
		this->getView().forAllRows( function );
		}

		template< typename Real,
		typename Device,
		typename Index,
		ElementsOrganization Organization,
		typename RealAllocator >
		template< typename Function >
		void
		DenseMatrix< Real, Device, Index, Organization, RealAllocator >::
		forAllRows( Function&& function ) const
		{
		this->getConsView().forAllRows( function );
		}

		template< typename Real,
		typename Device,
		typename Index,

src/TNL/Matrices/DenseMatrixView.h

+100 −0

Original line number	Diff line number	Diff line
		@@ -554,6 +554,106 @@ class DenseMatrixView : public MatrixView< Real, Device, Index >
		template< typename Function >
		void forAllElements( Function& function );

		/**
		* \brief Method for parallel iteration over matrix rows from interval [ \e begin, \e end).
		*
		* In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method
		* \ref DenseMatrix::forElements where more than one thread can be mapped to each row.
		*
		* \tparam Function is type of the lambda function.
		*
		* \param begin defines beginning of the range [ \e begin,\e end ) of rows to be processed.
		* \param end defines ending of the range [ \e begin, \e end ) of rows to be processed.
		* \param function is an instance of the lambda function to be called for each row.
		*
		* ```
		* auto function = [] __cuda_callable__ ( RowViewType& row ) mutable { ... };
		* ```
		*
		* \e RowViewType represents matrix row - see \ref TNL::Matrices::DenseMatrix::RowViewType.
		*
		* \par Example
		* \include Matrices/DenseMatrix/DenseMatrixExample_forRows.cpp
		* \par Output
		* \include DenseMatrixExample_forRows.out
		*/
		template< typename Function >
		void forRows( IndexType begin, IndexType end, Function&& function );

		/**
		* \brief Method for parallel iteration over matrix rows from interval [ \e begin, \e end) for constant instances.
		*
		* In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method
		* \ref DenseMatrixView::forElements where more than one thread can be mapped to each row.
		*
		* \tparam Function is type of the lambda function.
		*
		* \param begin defines beginning of the range [ \e begin,\e end ) of rows to be processed.
		* \param end defines ending of the range [ \e begin, \e end ) of rows to be processed.
		* \param function is an instance of the lambda function to be called for each row.
		*
		* ```
		* auto function = [] __cuda_callable__ ( RowViewType& row ) { ... };
		* ```
		*
		* \e RowViewType represents matrix row - see \ref TNL::Matrices::DenseMatrixView::RowViewType.
		*
		* \par Example
		* \include Matrices/DenseMatrix/DenseMatrixViewExample_forRows.cpp
		* \par Output
		* \include DenseMatrixViewExample_forRows.out
		*/
		template< typename Function >
		void forRows( IndexType begin, IndexType end, Function&& function ) const;

		/**
		* \brief Method for parallel iteration over all matrix rows.
		*
		* In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method
		* \ref DenseMatrixView::forAllElements where more than one thread can be mapped to each row.
		*
		* \tparam Function is type of the lambda function.
		*
		* \param function is an instance of the lambda function to be called for each row.
		*
		* ```
		* auto function = [] __cuda_callable__ ( RowViewType& row ) mutable { ... };
		* ```
		*
		* \e RowViewType represents matrix row - see \ref TNL::Matrices::DenseMatrixView::RowViewType.
		*
		* \par Example
		* \include Matrices/DenseMatrix/DenseMatrixViewExample_forRows.cpp
		* \par Output
		* \include DenseMatrixViewExample_forRows.out
		*/
		template< typename Function >
		void forAllRows( Function&& function );

		/**
		* \brief Method for parallel iteration over all matrix rows for constant instances.
		*
		* In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method
		* \ref DenseMatrixView::forAllElements where more than one thread can be mapped to each row.
		*
		* \tparam Function is type of the lambda function.
		*
		* \param function is an instance of the lambda function to be called for each row.
		*
		* ```
		* auto function = [] __cuda_callable__ ( RowViewType& row ) { ... };
		* ```
		*
		* \e RowViewType represents matrix row - see \ref TNL::Matrices::DenseMatrixView::RowViewType.
		*
		* \par Example
		* \include Matrices/DenseMatrix/DenseMatrixViewExample_forRows.cpp
		* \par Output
		* \include DenseMatrixViewExample_forRows.out
		*/
		template< typename Function >
		void forAllRows( Function&& function ) const;

		/**
		* \brief Method for sequential iteration over all matrix rows for constant instances.
		*

src/TNL/Matrices/DenseMatrixView.hpp

+60 −0

Original line number	Diff line number	Diff line
		@@ -385,6 +385,66 @@ forAllElements( Function& function )
		this->forElements( 0, this->getRows(), function );
		}

		template< typename Real,
		typename Device,
		typename Index,
		ElementsOrganization Organization >
		template< typename Function >
		void
		DenseMatrixView< Real, Device, Index, Organization >::
		forRows( IndexType begin, IndexType end, Function&& function )
		{
		auto values_view = this->values.getView();
		using SegmentViewType = typename SegmentsViewType::SegmentViewType;
		auto f = [=] __cuda_callable__ ( SegmentViewType& segmentView ) mutable {
		auto rowView = RowViewType( segmentView, values_view );
		function( rowView );
		};
		this->segments.forSegments( begin, end, f );
		}

		template< typename Real,
		typename Device,
		typename Index,
		ElementsOrganization Organization >
		template< typename Function >
		void
		DenseMatrixView< Real, Device, Index, Organization >::
		forRows( IndexType begin, IndexType end, Function&& function ) const
		{
		const auto values_view = this->values.getConstView();
		using SegmentViewType = typename SegmentsViewType::SegmentViewType;
		auto f = [=] __cuda_callable__ ( SegmentViewType&& segmentView ) mutable {
		const auto rowView = RowViewType( segmentView, values_view );
		function( rowView );
		};
		this->segments.forSegments( begin, end, f );
		}

		template< typename Real,
		typename Device,
		typename Index,
		ElementsOrganization Organization >
		template< typename Function >
		void
		DenseMatrixView< Real, Device, Index, Organization >::
		forAllRows( Function&& function )
		{
		this->forRows( 0, this->getRows(), function );
		}

		template< typename Real,
		typename Device,
		typename Index,
		ElementsOrganization Organization >
		template< typename Function >
		void
		DenseMatrixView< Real, Device, Index, Organization >::
		forAllRows( Function&& function ) const
		{
		this->forRows( 0, this->getRows(), function );
		}

		template< typename Real,
		typename Device,
		typename Index,