Fixing examples on forRows method in matrices. (74aa95aa) · Commits · TNL / tnl-dev

Documentation/Examples/Matrices/DenseMatrix/DenseMatrixExample_forRows.cpp

+14 −16

Original line number	Diff line number	Diff line
		#include <iostream>
		#include <TNL/Matrices/SparseMatrix.h>
		#include <TNL/Algorithms/ParallelFor.h>
		#include <TNL/Matrices/DenseMatrix.h>
		#include <TNL/Devices/Host.h>
		#include <TNL/Devices/Cuda.h>

		template< typename Device >
		void forRowsExample()
		{
		using MatrixType = TNL::Matrices::SparseMatrix< double, Device >;
		using RowViewType = typename MatrixType::RowViewType;
		MatrixType matrix( { 1, 2, 3, 4, 5, }, 5 );
		using MatrixType = TNL::Matrices::DenseMatrix< double, Device >;
		using RowView = typename MatrixType::RowView;
		MatrixType matrix( 5, 5 );

		auto f = [] __cuda_callable__ ( RowViewType& row ) mutable {
		for( int localIdx = 0;
		localIdx <= row.getRowIndex(); // This is important, some matrix formats may allocate more matrix elements
		localIdx++ ) // than we requested. These padding elements are processed here as well.
		// and so we cannot use row.getSize()
		{
		row.setValue( localIdx, row.getRowIndex() - localIdx + 1.0 );
		row.setColumnIndex( localIdx, localIdx );
		}
		auto f = [=] __cuda_callable__ ( RowView& row ) mutable {
		const int& rowIdx = row.getRowIndex();
		row.setElement( rowIdx, 10 * ( rowIdx + 1 ) );
		};
		matrix.forAllRows( f );

		/***
		* Set the matrix elements.
		*/
		matrix.forAllRows( f );
		std::cout << matrix << std::endl;
		}

		int main( int argc, char* argv[] )
		{
		std::cout << "Creating matrix on host: " << std::endl;
		std::cout << "Getting matrix rows on host: " << std::endl;
		forRowsExample< TNL::Devices::Host >();

		#ifdef HAVE_CUDA
		std::cout << "Creating matrix on CUDA device: " << std::endl;
		std::cout << "Getting matrix rows on CUDA device: " << std::endl;
		forRowsExample< TNL::Devices::Cuda >();
		#endif
		}

Documentation/Examples/Matrices/DenseMatrix/DenseMatrixViewExample_forRows.cpp

+14 −16

Original line number	Diff line number	Diff line
		#include <iostream>
		#include <TNL/Matrices/SparseMatrix.h>
		#include <TNL/Algorithms/ParallelFor.h>
		#include <TNL/Matrices/DenseMatrix.h>
		#include <TNL/Devices/Host.h>
		#include <TNL/Devices/Cuda.h>

		template< typename Device >
		void forRowsExample()
		{
		using MatrixType = TNL::Matrices::SparseMatrix< double, Device >;
		using RowViewType = typename MatrixType::RowViewType;
		MatrixType matrix( { 1, 2, 3, 4, 5, }, 5 );
		using MatrixType = TNL::Matrices::DenseMatrix< double, Device >;
		using RowView = typename MatrixType::RowView;
		MatrixType matrix( 5, 5 );
		auto view = matrix.getView();

		auto f = [] __cuda_callable__ ( RowViewType& row ) mutable {
		for( int localIdx = 0;
		localIdx <= row.getRowIndex(); // This is important, some matrix formats may allocate more matrix elements
		localIdx++ ) // than we requested. These padding elements are processed here as well.
		// and so we cannot use row.getSize()
		{
		row.setValue( localIdx, row.getRowIndex() - localIdx + 1.0 );
		row.setColumnIndex( localIdx, localIdx );
		}
		auto f = [=] __cuda_callable__ ( RowView& row ) mutable {
		const int& rowIdx = row.getRowIndex();
		row.setElement( rowIdx, 10 * ( rowIdx + 1 ) );
		};
		view.forAllRows( f );

		/***
		* Set the matrix elements.
		*/
		view.forAllRows( f );
		std::cout << matrix << std::endl;
		}

		int main( int argc, char* argv[] )
		{
		std::cout << "Creating matrix on host: " << std::endl;
		std::cout << "Getting matrix rows on host: " << std::endl;
		forRowsExample< TNL::Devices::Host >();

		#ifdef HAVE_CUDA
		std::cout << "Creating matrix on CUDA device: " << std::endl;
		std::cout << "Getting matrix rows on CUDA device: " << std::endl;
		forRowsExample< TNL::Devices::Cuda >();
		#endif
		}

Documentation/Examples/Matrices/LambdaMatrix/CMakeLists.txt

+1 −0

Original line number	Diff line number	Diff line
		@@ -4,6 +4,7 @@ set( COMMON_EXAMPLES
		LambdaMatrixExample_getNonzeroElementsCount
		LambdaMatrixExample_Laplace
		LambdaMatrixExample_Laplace_2
		LambdaMatrixExample_getRow
		LambdaMatrixExample_reduceRows
		LambdaMatrixExample_reduceAllRows
		LambdaMatrixExample_forElements

Documentation/Examples/Matrices/LambdaMatrix/LambdaMatrixExample_forRows.cpp

+55 −37

Original line number	Diff line number	Diff line
		@@ -8,55 +8,73 @@
		template< typename Device >
		void forRowsExample()
		{
		/**
		* Prepare lambda matrix of the following form:
		*
		* / 1 0 0 0 0 \
		* \| -2 1 -2 0 0 \|
		* \| 0 -2 1 -2 0 \|
		* \| 0 0 -2 1 -2 \|
		* \| 0 0 0 -2 1 \|
		* \ 0 0 0 0 1 /
		/***
		* Set matrix representing approximation of the Laplace operator on regular
		* grid using the finite difference method.
		*/

		int size = 5;
		auto rowLengths = [=] __cuda_callable__ ( const int rows, const int columns, const int rowIdx ) -> int {
		if( rowIdx == 0 \|\| rowIdx == size - 1 )
		const int gridSize( 4 );
		const int matrixSize = gridSize * gridSize;
		auto rowLengths = [=] __cuda_callable__ ( const int rows, const int columns, const int rowIdx ) -> int
		{
		const int gridRow = rowIdx / gridSize; // coordinates in the numerical grid
		const int gridColumn = rowIdx % gridSize;
		if( gridRow == 0 \|\| gridRow == gridSize - 1 \|\| // boundary grid node
		gridColumn == 0 \|\| gridColumn == gridSize - 1 )
		return 1;
		return 3;
		return 5;
		};

		auto matrixElements = [=] __cuda_callable__ ( const int rows, const int columns, const int rowIdx, const int localIdx, int& columnIdx, double& value) {
		if( rowIdx == 0 \|\| rowIdx == size -1 )
		const int gridRow = rowIdx / gridSize; // coordinates in the numerical grid
		const int gridColumn = rowIdx % gridSize;
		if( gridRow == 0 \|\| gridRow == gridSize - 1 \|\| // boundary grid node
		gridColumn == 0 \|\| gridColumn == gridSize - 1 )
		{
		columnIdx = rowIdx;
		value = 1.0;
		columnIdx = rowIdx; // diagonal element ....
		value = 1.0; // ... is set to 1
		}
		else
		else // interior grid node
		{
		columnIdx = rowIdx + localIdx - 1;
		value = ( columnIdx == rowIdx ) ? -2.0 : 1.0;
		switch( localIdx ) // set diagonal element to 4
		{ // and the others to -1
		case 0:
		columnIdx = rowIdx - gridSize;
		value = 1;
		break;
		case 1:
		columnIdx = rowIdx - 1;
		value = 1;
		break;
		case 2:
		columnIdx = rowIdx;
		value = -4;
		break;
		case 3:
		columnIdx = rowIdx + 1;
		value = 1;
		break;
		case 4:
		columnIdx = rowIdx + gridSize;
		value = 1;
		break;
		}
		}
		};
		auto matrix = TNL::Matrices::LambdaMatrixFactory< double, Device, int >::create(
		matrixSize, matrixSize, matrixElements, rowLengths );
		using MatrixType = decltype( matrix );

		using MatrixType = decltype( TNL::Matrices::LambdaMatrixFactory< double, Device, int >::create( matrixElements, rowLengths ) );
		MatrixType matrix( size, size, matrixElements, rowLengths );

		/**
		* Use the `forRows` method to copy the matrix elements to a dense matrix.
		*/
		TNL::Matrices::DenseMatrix< double, Device, int > denseMatrix( size, size );
		TNL::Matrices::DenseMatrix< double, Device > denseMatrix( matrixSize, matrixSize );
		denseMatrix.setValue( 0.0 );
		auto dense_view = denseMatrix.getView();
		auto f = [=] __cuda_callable__ ( const typename MatrixType::RowViewType& row ) mutable {
		auto f = [=] __cuda_callable__ ( const typename MatrixType::RowView& row ) mutable {
		auto dense_row = dense_view.getRow( row.getRowIndex() );
		for( int localIdx = 0; localIdx < row.getSize(); localIdx++ )
		dense_row.setElement( row.getColumnIndex( localIdx ), row.getValue( localIdx ) );
		};
		matrix.forAllRows( f );

		std::cout << "Lambda matrix looks as:" << std::endl << matrix << std::endl;
		std::cout << "Dense matrix looks as:" << std::endl << denseMatrix << std::endl;
		std::cout << "Laplace operator lambda matrix: " << std::endl << matrix << std::endl;
		std::cout << "Laplace operator dense matrix: " << std::endl << denseMatrix << std::endl;
		}

		int main( int argc, char* argv[] )

Documentation/Examples/Matrices/LambdaMatrix/LambdaMatrixExample_getRow.cpp

0 → 100644

+90 −0

Original line number	Diff line number	Diff line
		#include <iostream>
		#include <TNL/Matrices/LambdaMatrix.h>
		#include <TNL/Matrices/DenseMatrix.h>
		#include <TNL/Devices/Host.h>
		#include <TNL/Devices/Cuda.h>


		template< typename Device >
		void getRowExample()
		{
		/***
		* Set matrix representing approximation of the Laplace operator on regular
		* grid using the finite difference method.
		*/
		const int gridSize( 4 );
		const int matrixSize = gridSize * gridSize;
		auto rowLengths = [=] __cuda_callable__ ( const int rows, const int columns, const int rowIdx ) -> int
		{
		const int gridRow = rowIdx / gridSize; // coordinates in the numerical grid
		const int gridColumn = rowIdx % gridSize;
		if( gridRow == 0 \|\| gridRow == gridSize - 1 \|\| // boundary grid node
		gridColumn == 0 \|\| gridColumn == gridSize - 1 )
		return 1;
		return 5;
		};
		auto matrixElements = [=] __cuda_callable__ ( const int rows, const int columns, const int rowIdx, const int localIdx, int& columnIdx, double& value) {
		const int gridRow = rowIdx / gridSize; // coordinates in the numerical grid
		const int gridColumn = rowIdx % gridSize;
		if( gridRow == 0 \|\| gridRow == gridSize - 1 \|\| // boundary grid node
		gridColumn == 0 \|\| gridColumn == gridSize - 1 )
		{
		columnIdx = rowIdx; // diagonal element ....
		value = 1.0; // ... is set to 1
		}
		else // interior grid node
		{
		switch( localIdx ) // set diagonal element to 4
		{ // and the others to -1
		case 0:
		columnIdx = rowIdx - gridSize;
		value = 1;
		break;
		case 1:
		columnIdx = rowIdx - 1;
		value = 1;
		break;
		case 2:
		columnIdx = rowIdx;
		value = -4;
		break;
		case 3:
		columnIdx = rowIdx + 1;
		value = 1;
		break;
		case 4:
		columnIdx = rowIdx + gridSize;
		value = 1;
		break;
		}
		}
		};
		auto matrix = TNL::Matrices::LambdaMatrixFactory< double, Device, int >::create(
		matrixSize, matrixSize, matrixElements, rowLengths );
		using MatrixType = decltype( matrix );

		TNL::Matrices::DenseMatrix< double, Device > denseMatrix( matrixSize, matrixSize );
		denseMatrix.setValue( 0.0 );
		auto dense_view = denseMatrix.getView();
		auto f = [=] __cuda_callable__ ( const int& rowIdx ) mutable {
		auto row = matrix.getRow( rowIdx );
		auto dense_row = dense_view.getRow( rowIdx );
		for( int localIdx = 0; localIdx < row.getSize(); localIdx++ )
		dense_row.setElement( row.getColumnIndex( localIdx ), row.getValue( localIdx ) );
		};
		TNL::Algorithms::ParallelFor< Device >::exec( 0, matrixSize, f );

		std::cout << "Laplace operator lambda matrix: " << std::endl << matrix << std::endl;
		std::cout << "Laplace operator dense matrix: " << std::endl << denseMatrix << std::endl;
		}

		int main( int argc, char* argv[] )
		{
		std::cout << "Running example on CPU ... " << std::endl;
		getRowExample< TNL::Devices::Host >();

		#ifdef HAVE_CUDA
		std::cout << "Running example on CUDA GPU ... " << std::endl;
		getRowExample< TNL::Devices::Cuda >();
		#endif
		}