Commit f4bab3c9 authored by Lukas Cejka's avatar Lukas Cejka Committed by Tomáš Oberhuber
Browse files

Implemented fix for case when roundToMultiple would give a number smaller than... 

Implemented fix for case when roundToMultiple would give a number smaller than the number of rows, thus causing index to be out of bounds on CUDA
parent 4e1309ce

src/TNL/Matrices/EllpackSymmetricGraph_impl.h

+17 −3
Original line number Diff line number Diff line
@@ -73,11 +73,25 @@ void EllpackSymmetricGraph< Real, Device, Index >::setDimensions( const IndexTyp 
   TNL_ASSERT( rows > 0 && columns > 0,

              std::cerr << "rows = " << rows

                   << " columns = " << columns << std::endl );

   

   this->rows = rows;

   this->columns = columns;

   

   if( std::is_same< DeviceType, Devices::Cuda >::value )

      this->alignedRows = roundToMultiple( columns, Cuda::getWarpSize() );

   {

       this->alignedRows = roundToMultiple( columns, Devices::Cuda::getWarpSize() );

       

       if( this->rows - this->alignedRows > 0 )

       {

           IndexType missingRows = this->rows - this->alignedRows;

           missingRows = roundToMultiple( missingRows, Devices::Cuda::getWarpSize() );

           this->alignedRows +=  missingRows;

           

//           this->alignedRows += roundToMultiple( this->rows - this->alignedRows, Devices::Cuda::getWarpSize() );

       }

   }

   else this->alignedRows = rows;

   

   if( this->rowLengths != 0 )

       allocateElements();

}