Commit ae1def7a authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Increased bufferRowsCount in SparseMatrix from 128 to 4096 

Buffering with a small value is very slow when profiling anything in nvpp.
parent dcbd2733

src/TNL/Matrices/SparseMatrix.hpp

+2 −2
Original line number Diff line number Diff line
@@ -904,7 +904,7 @@ operator=( const DenseMatrix< Real_, Device_, Index_, Organization, RealAllocato 
   else

   {

      const IndexType maxRowLength = matrix.getColumns();

      const IndexType bufferRowsCount( 128 );

      const IndexType bufferRowsCount( 4096 );

      const size_t bufferSize = bufferRowsCount * maxRowLength;

      Containers::Vector< RHSRealType, RHSDeviceType, RHSIndexType, RHSRealAllocatorType > matrixValuesBuffer( bufferSize );

      Containers::Vector< RealType, DeviceType, IndexType, RealAllocatorType > thisValuesBuffer( bufferSize );
@@ -1017,7 +1017,7 @@ operator=( const RHSMatrix& matrix ) 
   else

   {

      const IndexType maxRowLength = max( rowCapacities );

      const IndexType bufferRowsCount( 128 );

      const IndexType bufferRowsCount( 4096 );

      const size_t bufferSize = bufferRowsCount * maxRowLength;

      Containers::Vector< RHSRealType, RHSDeviceType, RHSIndexType, RHSRealAllocatorType > matrixValuesBuffer( bufferSize );

      Containers::Vector< RHSIndexType, RHSDeviceType, RHSIndexType > matrixColumnsBuffer( bufferSize );