Removing globalIdx from Fetch lambda in SparseMatrixView::rowsReduction. (9578471a) · Commits · TNL / tnl-dev

src/TNL/Matrices/DenseMatrixView.hpp

+4 −0

Original line number	Diff line number	Diff line
		@@ -40,8 +40,12 @@ DenseMatrixView( const IndexType rows,
		const ValuesViewType& values )
		: MatrixView< Real, Device, Index >( rows, columns, values )
		{
		#ifdef __CUDA_ARCH__
		TNL_ASSERT_EQ( values.getSize(), this->getAllocatedElementsCount(), "Number of matrix elements does not agree with matrix dimensions." );
		#else
		if( values.getSize() != this->getAllocatedElementsCount() )
		throw( std::logic_error( "Number of matrix elements does not agree with matrix dimensions." ) );
		#endif

		SegmentsType a( rows, columns );
		segments = a.getView();

src/TNL/Matrices/SparseMatrixView.hpp

+4 −4

Original line number	Diff line number	Diff line
		@@ -121,7 +121,7 @@ getCompressedRowLengths( Vector& rowLengths ) const
		details::set_size_if_resizable( rowLengths, this->getRows() );
		rowLengths = 0;
		auto rowLengths_view = rowLengths.getView();
		auto fetch = [] __cuda_callable__ ( IndexType row, IndexType column, IndexType globalIdx, const RealType& value ) -> IndexType {
		auto fetch = [] __cuda_callable__ ( IndexType row, IndexType column, const RealType& value ) -> IndexType {
		return ( value != 0.0 );
		};
		auto keep = [=] __cuda_callable__ ( const IndexType rowIdx, const IndexType value ) mutable {
		@@ -448,14 +448,14 @@ rowsReduction( IndexType begin, IndexType end, Fetch& fetch, const Reduce& reduc
		const auto columns_view = this->columnIndexes.getConstView();
		const auto values_view = this->values.getConstView();
		const IndexType paddingIndex_ = this->getPaddingIndex();
		auto fetch_ = [=] __cuda_callable__ ( IndexType rowIdx, IndexType localIdx, IndexType globalIdx, bool& compute ) mutable -> decltype( fetch( IndexType(), IndexType(), IndexType(), RealType() ) ) {
		auto fetch_ = [=] __cuda_callable__ ( IndexType rowIdx, IndexType localIdx, IndexType globalIdx, bool& compute ) mutable -> decltype( fetch( IndexType(), IndexType(), RealType() ) ) {
		IndexType columnIdx = columns_view[ globalIdx ];
		if( columnIdx != paddingIndex_ )
		{
		if( isBinary() )
		return fetch( rowIdx, columnIdx, globalIdx, 1 );
		return fetch( rowIdx, columnIdx, 1 );
		else
		return fetch( rowIdx, columnIdx, globalIdx, values_view[ globalIdx ] );
		return fetch( rowIdx, columnIdx, values_view[ globalIdx ] );
		}
		return zero;
		};

src/UnitTests/Matrices/BinarySparseMatrixTest.hpp

+2 −2

Original line number	Diff line number	Diff line
		@@ -948,7 +948,7 @@ void test_RowsReduction()
		// Compute number of non-zero elements in rows.
		typename Matrix::RowsCapacitiesType rowLengths( rows );
		auto rowLengths_view = rowLengths.getView();
		auto fetch = [] __cuda_callable__ ( IndexType row, IndexType column, IndexType globalIdx, const RealType& value ) -> IndexType {
		auto fetch = [] __cuda_callable__ ( IndexType row, IndexType column, const RealType& value ) -> IndexType {
		return ( value != 0.0 );
		};
		auto keep = [=] __cuda_callable__ ( const IndexType rowIdx, const IndexType value ) mutable {
		@@ -963,7 +963,7 @@ void test_RowsReduction()
		// Compute max norm
		TNL::Containers::Vector< RealType, DeviceType, IndexType > rowSums( rows );
		auto rowSums_view = rowSums.getView();
		auto max_fetch = [] __cuda_callable__ ( IndexType row, IndexType column, IndexType globalIdx, const RealType& value ) -> IndexType {
		auto max_fetch = [] __cuda_callable__ ( IndexType row, IndexType column, const RealType& value ) -> IndexType {
		return abs( value );
		};
		auto max_keep = [=] __cuda_callable__ ( const IndexType rowIdx, const IndexType value ) mutable {

src/UnitTests/Matrices/SparseMatrixTest.hpp

+2 −2

Original line number	Diff line number	Diff line
		@@ -1374,7 +1374,7 @@ void test_RowsReduction()
		// Compute number of non-zero elements in rows.
		typename Matrix::RowsCapacitiesType rowLengths( rows );
		auto rowLengths_view = rowLengths.getView();
		auto fetch = [] __cuda_callable__ ( IndexType row, IndexType column, IndexType globalIdx, const RealType& value ) -> IndexType {
		auto fetch = [] __cuda_callable__ ( IndexType row, IndexType column, const RealType& value ) -> IndexType {
		return ( value != 0.0 );
		};
		auto keep = [=] __cuda_callable__ ( const IndexType rowIdx, const IndexType value ) mutable {
		@@ -1389,7 +1389,7 @@ void test_RowsReduction()
		// Compute max norm
		TNL::Containers::Vector< RealType, DeviceType, IndexType > rowSums( rows );
		auto rowSums_view = rowSums.getView();
		auto max_fetch = [] __cuda_callable__ ( IndexType row, IndexType column, IndexType globalIdx, const RealType& value ) -> IndexType {
		auto max_fetch = [] __cuda_callable__ ( IndexType row, IndexType column, const RealType& value ) -> IndexType {
		return abs( value );
		};
		auto max_keep = [=] __cuda_callable__ ( const IndexType rowIdx, const IndexType value ) mutable {

src/UnitTests/Matrices/SymmetricSparseMatrixTest.hpp

+1 −1

Original line number	Diff line number	Diff line
		@@ -898,7 +898,7 @@ void test_RowsReduction()
		typename Matrix::RowsCapacitiesType rowLengths_true( { 1, 1, 4, 4, 4, 4, 1, 1 } );
		auto rowLengths_view = rowLengths.getView();
		rowLengths_view = 0;
		auto fetch = [=] __cuda_callable__ ( IndexType row, IndexType column, IndexType globalIdx, const RealType& value ) mutable -> IndexType {
		auto fetch = [=] __cuda_callable__ ( IndexType row, IndexType column, const RealType& value ) mutable -> IndexType {
		if( value != 0.0 && row != column)
		TNL::Algorithms::AtomicOperations< DeviceType >::add( rowLengths_view[ column ], ( IndexType ) 1 );
		return ( value != 0.0 );