oprava COO a Cusparse (46717493) · Commits · TNL / tnl-dev

src/implementation/matrices/tnlCOOMatrix_impl.h

+37 −4

Original line number	Diff line number	Diff line
		@@ -11,7 +11,8 @@ template< typename Real,
		typename Index >
		tnlCOOMatrix< Real, Device, Index >::tnlCOOMatrix()
		:cudaWarpSize( 32 ),
		numberOfUsedValues( 0 )
		numberOfUsedValues( 0 ),
		appendMode( true )
		{
		};

		@@ -89,8 +90,14 @@ template< typename Real,
		typename Index >
		void tnlCOOMatrix< Real, Device, Index >::getRowLengths(tnlVector< IndexType, DeviceType, IndexType >& rowLengthsVector) const
		{
		IndexType rowLength;
		for(IndexType row = 0; row < this->getRows(); row++)
		rowLengthsVector.setElement(row, this->getRowLength(row));
		rowLengthsVector.setElement(row, 0);
		for(IndexType elementPtr = 0; elementPtr < this->values.getSize(); elementPtr++)
		{
		rowLength = rowLengthsVector.getElement(this->rowIndexes.getElement(elementPtr));
		rowLengthsVector.setElement(this->rowIndexes.getElement(elementPtr), rowLength++);
		}
		}

		template< typename Real,
		@@ -112,7 +119,10 @@ bool tnlCOOMatrix< Real, Device, Index >::setElement( const IndexType row,
		const IndexType column,
		const RealType& value )
		{
		if( this->appendMode )
		return this->addElement( row, column, value, 1.0 );
		else
		return this->appendElement( row, column, value );
		}

		template< typename Real,
		@@ -129,6 +139,8 @@ bool tnlCOOMatrix< Real, Device, Index >::addElement( const IndexType row,
		<< " column = " << column
		<< " this->rows = " << this->rows
		<< " this->columns = " << this->columns );
		if( appendMode )
		return this->appendElement( row, column, value );

		IndexType endPtr = this->getNumberOfUsedValues();
		for(IndexType elementPtr = 0; elementPtr < endPtr; elementPtr++)
		@@ -150,6 +162,25 @@ bool tnlCOOMatrix< Real, Device, Index >::addElement( const IndexType row,
		return false;
		}

		template< typename Real,
		typename Device,
		typename Index >
		bool tnlCOOMatrix< Real, Device, Index >::appendElement( const IndexType row,
		const IndexType column,
		const RealType& value )
		{
		if( !this->getNumberOfUsedValues < this->values.getSize() )
		return false;
		else
		{
		this->rowIndexes.setElement( this->getNumberOfUsedValues(), row );
		this->columnIndexes.setElement( this->getNumberOfUsedValues(), column );
		this->values.setElement( this->getNumberOfUsedValues(), value );
		this->numberOfUsedValues++;
		}
		return true;
		}

		template< typename Real,
		typename Device,
		typename Index >
		@@ -236,6 +267,8 @@ template< typename InVector,
		void tnlCOOMatrix< Real, Device, Index >::vectorProductHost(const InVector& inVector,
		OutVector& outVector) const
		{
		for(IndexType row = 0; row < this->getRows(); row++)
		outVector[ row ] = 0;
		for(IndexType i = 0; i < this->values.getSize(); i++)
		outVector[ this->rowIndexes.getElement(i) ] = this->values.getElement(i)*inVector[ this->columnIndexes.getElement(i) ];
		}

src/matrices/tnlCOOMatrix.h

+8 −0

Original line number	Diff line number	Diff line
		/* u addElement by nejspis melo byt realokovani pole jinak se asi prvek, ktery
		* na danem miste nebyl pridat neda, leda by se puvodne naalokovalo pole o neco vetsi
		*
		* u getRowLengths dat jeden cyklus co projede vsechny prvky a nastavi rovnou cele pole
		*/
		#ifndef TNLCOOMATRIX_H_
		#define TNLCOOMATRIX_H_
		@@ -51,6 +53,10 @@ public:
		const RealType& value,
		const RealType& thisElementMultiplicator = 1.0);

		bool appendElement( const IndexType row,
		const IndexType column,
		const RealType& value);

		bool setRow(const IndexType row,
		const IndexType* columns,
		const RealType* values,
		@@ -106,6 +112,8 @@ private:
		IndexType numberOfUsedValues;

		int cudaWarpSize;

		bool appendMode;
		};

		#include <implementation/matrices/tnlCOOMatrix_impl.h>

tests/benchmarks/tnlCusparseCSRMatrix.h

+7 −5

Original line number	Diff line number	Diff line
		@@ -110,18 +110,19 @@ class tnlCusparseCSRMatrix< double > : public tnlCusparseCSRMatrixBase< double >
		{
		tnlAssert( matrix, );
		#ifdef HAVE_CUDA
		/cusparseDcsrmv( ( this->cusparseHandle ),
		cusparseDcsrmv( *( this->cusparseHandle ),
		CUSPARSE_OPERATION_NON_TRANSPOSE,
		this->matrix->getRows(),
		this->matrix->getColumns(),
		this->matrix->values.getSize(),
		1.0,
		this->matrixDescriptor,
		this->matrix->values.getData(),
		this->matrix->rowPointers.getData(),
		this->matrix->columnIndexes.getData(),
		inVector.getData(),
		1.0,
		outVector.getData() );*/
		outVector.getData() );
		#endif
		}
		};
		@@ -138,18 +139,19 @@ class tnlCusparseCSRMatrix< float > : public tnlCusparseCSRMatrixBase< float >
		{
		tnlAssert( matrix, );
		#ifdef HAVE_CUDA
		/cusparseScsrmv( ( this->cusparseHandle ),
		cusparseScsrmv( *( this->cusparseHandle ),
		CUSPARSE_OPERATION_NON_TRANSPOSE,
		this->matrix->getRows(),
		this->matrix->getColumns(),
		this->matrix->values.getSize(),
		1.0,
		this->matrixDescriptor,
		this->matrix->values.getData(),
		this->matrix->rowPointers.getData(),
		this->matrix->columnIndexes.getData(),
		inVector.getData(),
		0,
		outVector.getData() );*/
		outVector.getData() );
		#endif
		}
		};