Commit 5f98eede authored by Tomáš Oberhuber's avatar Tomáš Oberhuber
Browse files

Deleted DenseMatrix[View]::getRowVectorProduct and added rows range to vectorProduct method.

parent 344b5e07

src/TNL/Matrices/DenseMatrix.h

+17 −25
Original line number Diff line number Diff line
@@ -331,9 +331,10 @@ class DenseMatrix : public Matrix< Real, Device, Index > 
      /**

       * \brief Sets element at given \e row and \e column to given \e value.

       * 

       * This method can be called only from the host system (CPU) no matter

       * where the matrix is allocated. If the matrix is allocated in GPU device

       * this methods transfer values of each matrix element separately and so the

       * This method can be called from the host system (CPU) no matter

       * where the matrix is allocated. If the matrix is allocated on GPU this method

       * can be called even from device kernels. If the matrix is allocated in GPU device

       * this method is called from CPU, it transfers values of each matrix element separately and so the

       * performance is very low. For higher performance see. \ref DenseMatrix::getRow

       * or \ref DenseMatrix::forRows and \ref DenseMatrix::forAllRows.

       *
@@ -354,9 +355,10 @@ class DenseMatrix : public Matrix< Real, Device, Index > 
      /**

       * \brief Add element at given \e row and \e column to given \e value.

       * 

       * This method can be called only from the host system (CPU) no matter

       * where the matrix is allocated. If the matrix is allocated in GPU device

       * this methods transfer values of each matrix element separately and so the

       * This method can be called from the host system (CPU) no matter

       * where the matrix is allocated. If the matrix is allocated on GPU this method

       * can be called even from device kernels. If the matrix is allocated in GPU device

       * this method is called from CPU, it transfers values of each matrix element separately and so the

       * performance is very low. For higher performance see. \ref DenseMatrix::getRow

       * or \ref DenseMatrix::forRows and \ref DenseMatrix::forAllRows.

       *
@@ -375,9 +377,10 @@ class DenseMatrix : public Matrix< Real, Device, Index > 
      /**

       * \brief Returns value of matrix element at position given by its row and column index.

       * 

       * This method can be called only from the host system (CPU) no matter

       * where the matrix is allocated. If the matrix is allocated in GPU device

       * this methods transfer values of each matrix element separately and so the

       * This method can be called from the host system (CPU) no matter

       * where the matrix is allocated. If the matrix is allocated on GPU this method

       * can be called even from device kernels. If the matrix is allocated in GPU device

       * this method is called from CPU, it transfers values of each matrix element separately and so the

       * performance is very low. For higher performance see. \ref DenseMatrix::getRow

       * or \ref DenseMatrix::forRows and \ref DenseMatrix::forAllRows.

       *
@@ -518,21 +521,6 @@ class DenseMatrix : public Matrix< Real, Device, Index > 
      template< typename Function >

      void forAllRows( Function& function );



      /**

       * \brief This method computes scalar product of given vector and one 

       *  row of the matrix.

       * 

       * \tparam Vector is type of input vector. It can be \ref Vector,

       *     \ref VectorView, \ref Array, \ref ArraView or similar container.

       * \param row is index of the row used for the scalar product.

       * \param vector is the input vector.

       * \return result of the matrix row and vector product.

       */

      template< typename Vector >

      __cuda_callable__

      typename Vector::RealType rowVectorProduct( const IndexType row,

                                                  const Vector& vector ) const;



      /**

       * \brief Computes product of matrix and vector.

       *
@@ -546,7 +534,11 @@ class DenseMatrix : public Matrix< Real, Device, Index > 
       */

      template< typename InVector, typename OutVector >

      void vectorProduct( const InVector& inVector,

                          OutVector& outVector ) const;

                          OutVector& outVector,

                          const RealType& matrixMultiplicator = 1.0,

                          const RealType& outVectorMultiplicator = 0.0,

                          const IndexType firstRow = 0,

                          const IndexType lastRow = 0 ) const;



      template< typename Matrix >

      void addMatrix( const Matrix& matrix,

src/TNL/Matrices/DenseMatrix.hpp

+7 −16
Original line number Diff line number Diff line
@@ -416,20 +416,6 @@ forAllRows( Function& function ) 
   this->forRows( 0, this->getRows(), function );

}



template< typename Real,

          typename Device,

          typename Index,

          bool RowMajorOrder,

          typename RealAllocator >

   template< typename Vector >

__cuda_callable__

typename Vector::RealType 

DenseMatrix< Real, Device, Index, RowMajorOrder, RealAllocator >::

rowVectorProduct( const IndexType row, const Vector& vector ) const

{

   return this->view.rowVectorProduct( row, vector );

}



template< typename Real,

          typename Device,

          typename Index,
@@ -439,9 +425,14 @@ template< typename Real, 
             typename OutVector >

void

DenseMatrix< Real, Device, Index, RowMajorOrder, RealAllocator >::

vectorProduct( const InVector& inVector, OutVector& outVector ) const

vectorProduct( const InVector& inVector,

               OutVector& outVector,

               const RealType& matrixMultiplicator,

               const RealType& outVectorMultiplicator,

               const IndexType firstRow,

               const IndexType lastRow ) const

{

   this->view.vectorProduct( inVector, outVector );

   this->view.vectorProduct( inVector, outVector, matrixMultiplicator, outVectorMultiplicator, firstRow, lastRow );

}



template< typename Real,

src/TNL/Matrices/DenseMatrixView.h

+5 −6
Original line number Diff line number Diff line
@@ -141,14 +141,13 @@ class DenseMatrixView : public MatrixView< Real, Device, Index > 
      template< typename Function >

      void forAllRows( Function& function );



      template< typename Vector >

      __cuda_callable__

      typename Vector::RealType rowVectorProduct( const IndexType row,

                                                  const Vector& vector ) const;



      template< typename InVector, typename OutVector >

      void vectorProduct( const InVector& inVector,

                          OutVector& outVector ) const;

                          OutVector& outVector,

                          const RealType& matrixMultiplicator = 1.0,

                          const RealType& outVectorMultiplicator = 0.0,

                          const IndexType firstRow = 0,

                          IndexType lastRow = 0 ) const;



      template< typename Matrix >

      void addMatrix( const Matrix& matrix,

src/TNL/Matrices/DenseMatrixView.hpp

+10 −20
Original line number Diff line number Diff line
@@ -367,23 +367,6 @@ forAllRows( Function& function ) 
   this->forRows( 0, this->getRows(), function );

}



template< typename Real,

          typename Device,

          typename Index,

          bool RowMajorOrder >

   template< typename Vector >

__cuda_callable__

typename Vector::RealType

DenseMatrixView< Real, Device, Index, RowMajorOrder >::

rowVectorProduct( const IndexType row, const Vector& vector ) const

{

   RealType sum( 0.0 );

   // TODO: Fix this

   //for( IndexType column = 0; column < this->getColumns(); column++ )

   //   sum += this->getElementFast( row, column ) * vector[ column ];

   return sum;

}



template< typename Real,

          typename Device,

          typename Index,
@@ -392,7 +375,12 @@ template< typename Real, 
             typename OutVector >

void

DenseMatrixView< Real, Device, Index, RowMajorOrder >::

vectorProduct( const InVector& inVector, OutVector& outVector ) const

vectorProduct( const InVector& inVector,

               OutVector& outVector,

               const RealType& matrixMultiplicator,

               const RealType& outVectorMultiplicator,

               const IndexType firstRow,

               IndexType lastRow ) const

{

   TNL_ASSERT_EQ( this->getColumns(), inVector.getSize(), "Matrix columns count differs with input vector size." );

   TNL_ASSERT_EQ( this->getRows(), outVector.getSize(), "Matrix rows count differs with output vector size." );
@@ -400,13 +388,15 @@ vectorProduct( const InVector& inVector, OutVector& outVector ) const 
   const auto inVectorView = inVector.getConstView();

   auto outVectorView = outVector.getView();

   const auto valuesView = this->values.getConstView();

   if( lastRow == 0 )

      lastRow = this->getRows();

   auto fetch = [=] __cuda_callable__ ( IndexType row, IndexType column, IndexType offset, bool& compute ) -> RealType {

      return valuesView[ offset ] * inVectorView[ column ];

   };

   auto keeper = [=] __cuda_callable__ ( IndexType row, const RealType& value ) mutable {

      outVectorView[ row ] = value;

      outVectorView[ row ] = matrixMultiplicator * value + outVectorMultiplicator * outVectorView[ row ];

   };

   this->segments.segmentsReduction( 0, this->getRows(), fetch, std::plus<>{}, keeper, ( RealType ) 0.0 );

   this->segments.segmentsReduction( firstRow, lastRow, fetch, std::plus<>{}, keeper, ( RealType ) 0.0 );

}



template< typename Real,