Methods DenseMatrix[View]::get/set/addElement are __cuda_callble__. (344b5e07) · Commits · TNL / tnl-dev

src/TNL/Matrices/DenseMatrix.h

+3 −0

Original line number	Diff line number	Diff line
		@@ -346,6 +346,7 @@ class DenseMatrix : public Matrix< Real, Device, Index >
		* \par Output
		* \include DenseMatrixExample_setElement.out
		*/
		__cuda_callable__
		void setElement( const IndexType row,
		const IndexType column,
		const RealType& value );
		@@ -365,6 +366,7 @@ class DenseMatrix : public Matrix< Real, Device, Index >
		* \param thisElementMultiplicator is multiplicator the original matrix element
		* value is multiplied by before addition of given e value.
		*/
		__cuda_callable__
		void addElement( const IndexType row,
		const IndexType column,
		const RealType& value,
		@@ -384,6 +386,7 @@ class DenseMatrix : public Matrix< Real, Device, Index >
		*
		* \return value of given matrix element.
		*/
		__cuda_callable__
		Real getElement( const IndexType row,
		const IndexType column ) const;

src/TNL/Matrices/DenseMatrix.hpp

+4 −3

Original line number	Diff line number	Diff line
		@@ -233,6 +233,7 @@ DenseMatrix< Real, Device, Index, RowMajorOrder, RealAllocator >::
		reset()
		{
		Matrix< Real, Device, Index >::reset();
		this->segments.reset();
		}

		template< typename Real,
		@@ -300,7 +301,7 @@ template< typename Real,
		typename Index,
		bool RowMajorOrder,
		typename RealAllocator >
		void
		__cuda_callable__ void
		DenseMatrix< Real, Device, Index, RowMajorOrder, RealAllocator >::
		setElement( const IndexType row,
		const IndexType column,
		@@ -314,7 +315,7 @@ template< typename Real,
		typename Index,
		bool RowMajorOrder,
		typename RealAllocator >
		void
		__cuda_callable__ void
		DenseMatrix< Real, Device, Index, RowMajorOrder, RealAllocator >::
		addElement( const IndexType row,
		const IndexType column,
		@@ -329,7 +330,7 @@ template< typename Real,
		typename Index,
		bool RowMajorOrder,
		typename RealAllocator >
		Real
		__cuda_callable__ Real
		DenseMatrix< Real, Device, Index, RowMajorOrder, RealAllocator >::
		getElement( const IndexType row,
		const IndexType column ) const

src/TNL/Matrices/DenseMatrixView.h

+3 −2

Original line number	Diff line number	Diff line
		@@ -91,8 +91,6 @@ class DenseMatrixView : public MatrixView< Real, Device, Index >

		IndexType getNonzeroElementsCount() const;

		void reset();

		__cuda_callable__
		const RowView getRow( const IndexType& rowIdx ) const;

		@@ -110,15 +108,18 @@ class DenseMatrixView : public MatrixView< Real, Device, Index >
		const Real& operator()( const IndexType row,
		const IndexType column ) const;

		__cuda_callable__
		void setElement( const IndexType row,
		const IndexType column,
		const RealType& value );

		__cuda_callable__
		void addElement( const IndexType row,
		const IndexType column,
		const RealType& value,
		const RealType& thisElementMultiplicator = 1.0 );

		__cuda_callable__
		Real getElement( const IndexType row,
		const IndexType column ) const;

src/TNL/Matrices/DenseMatrixView.hpp

+3 −14

Original line number	Diff line number	Diff line
		@@ -168,17 +168,6 @@ getNonzeroElementsCount() const
		return Algorithms::Reduction< DeviceType >::reduce( this->values.getSize(), std::plus<>{}, fetch, 0 );
		}

		template< typename Real,
		typename Device,
		typename Index,
		bool RowMajorOrder >
		void
		DenseMatrixView< Real, Device, Index, RowMajorOrder >::
		reset()
		{
		Matrix< Real, Device, Index >::reset();
		}

		template< typename Real,
		typename Device,
		typename Index,
		@@ -250,7 +239,7 @@ template< typename Real,
		typename Device,
		typename Index,
		bool RowMajorOrder >
		void
		__cuda_callable__ void
		DenseMatrixView< Real, Device, Index, RowMajorOrder >::
		setElement( const IndexType row,
		const IndexType column,
		@@ -263,7 +252,7 @@ template< typename Real,
		typename Device,
		typename Index,
		bool RowMajorOrder >
		void
		__cuda_callable__ void
		DenseMatrixView< Real, Device, Index, RowMajorOrder >::
		addElement( const IndexType row,
		const IndexType column,
		@@ -283,7 +272,7 @@ template< typename Real,
		typename Device,
		typename Index,
		bool RowMajorOrder >
		Real
		__cuda_callable__ Real
		DenseMatrixView< Real, Device, Index, RowMajorOrder >::
		getElement( const IndexType row,
		const IndexType column ) const

src/UnitTests/Matrices/DenseMatrixTest.h

+177 −161

Original line number	Diff line number	Diff line
		@@ -8,15 +8,17 @@

		/* See Copyright Notice in tnl/Copyright */

		#include <iostream>
		#include <functional>
		#include <TNL/Devices/Host.h>
		#include <TNL/Matrices/Matrix.h>
		#include <TNL/Matrices/DenseMatrix.h>
		#include <TNL/Containers/Array.h>

		#include <TNL/Containers/Vector.h>
		#include <TNL/Containers/VectorView.h>
		#include <TNL/Algorithms/ParallelFor.h>
		#include <TNL/Algorithms/Reduction.h>
		#include <TNL/Math.h>
		#include <iostream>

		using Dense_host_float = TNL::Matrices::DenseMatrix< float, TNL::Devices::Host, int >;
		using Dense_host_int = TNL::Matrices::DenseMatrix< int, TNL::Devices::Host, int >;
		@@ -380,6 +382,7 @@ void test_SetElement()
		using RealType = typename Matrix::RealType;
		using DeviceType = typename Matrix::DeviceType;
		using IndexType = typename Matrix::IndexType;

		/*
		* Sets up the following 5x5 dense matrix:
		*
		@@ -392,9 +395,7 @@ void test_SetElement()
		const IndexType rows = 5;
		const IndexType cols = 5;

		Matrix m;
		m.reset();
		m.setDimensions( rows, cols );
		Matrix m( rows, cols );

		RealType value = 1;
		for( IndexType i = 0; i < rows; i++ )
		@@ -430,6 +431,22 @@ void test_SetElement()
		EXPECT_EQ( m.getElement( 4, 2 ), 23 );
		EXPECT_EQ( m.getElement( 4, 3 ), 24 );
		EXPECT_EQ( m.getElement( 4, 4 ), 25 );

		TNL::Containers::Vector< RealType, DeviceType, IndexType > v( m.getRows(), 0 );
		auto v_view = v.getView();
		auto m_view = m.getView();
		auto f1 = [=] __cuda_callable__ ( IndexType i ) mutable {
		v_view[ i ] = m_view.getElement( i, i );
		};
		TNL::Algorithms::ParallelFor< DeviceType >::exec( ( IndexType ) 0, m.getRows(), f1 );

		for( IndexType i = 0; i < m.getRows(); i++ )
		EXPECT_EQ( v.getElement( i ), m.getElement( i, i ) );
		auto fetch = [=] __cuda_callable__ ( IndexType i ) -> bool {
		return ( v_view[ i ] == m_view.getElement( i, i ) );
		};
		EXPECT_TRUE( TNL::Algorithms::Reduction< DeviceType >::reduce( m.getRows(), std::logical_and<>{}, fetch, true ) );

		}

		template< typename Matrix >
		@@ -438,6 +455,7 @@ void test_AddElement()
		using RealType = typename Matrix::RealType;
		using DeviceType = typename Matrix::DeviceType;
		using IndexType = typename Matrix::IndexType;

		/*
		* Sets up the following 6x5 dense matrix:
		*
		@@ -451,9 +469,7 @@ void test_AddElement()
		const IndexType rows = 6;
		const IndexType cols = 5;

		Matrix m;
		m.reset();
		m.setDimensions( rows, cols );
		Matrix m( rows, cols );

		RealType value = 1;
		for( IndexType i = 0; i < rows; i++ )