Commit 8c17519f authored by Jakub Klinkovský's avatar Jakub Klinkovský Committed by Tomáš Oberhuber
Browse files

Started porting DistributedMatrix to segments

parent c6d1173d

src/TNL/Matrices/DistributedMatrix.h

+11 −12
Original line number Diff line number Diff line
@@ -14,7 +14,6 @@ 


#include <type_traits>



#include <TNL/Matrices/Legacy/SparseRow.h>

#include <TNL/Communicators/MpiCommunicator.h>

#include <TNL/Containers/Subrange.h>

#include <TNL/Containers/DistributedVector.h>
@@ -56,8 +55,8 @@ public: 


   using CompressedRowLengthsVector = Containers::DistributedVector< IndexType, DeviceType, IndexType, CommunicatorType >;



   using MatrixRow = Matrices::Legacy::SparseRow< RealType, IndexType >;

   using ConstMatrixRow = Matrices::Legacy::SparseRow< std::add_const_t< RealType >, std::add_const_t< IndexType > >;

   using MatrixRow = typename Matrix::RowView;

   using ConstMatrixRow = typename Matrix::ConstRowView;



   template< typename _Real = RealType,

             typename _Device = DeviceType,
@@ -125,16 +124,16 @@ public: 
   RealType getElementFast( IndexType row,

                            IndexType column ) const;



   __cuda_callable__

   bool setRowFast( IndexType row,

                    const IndexType* columnIndexes,

                    const RealType* values,

                    IndexType elements );

//   __cuda_callable__

//   bool setRowFast( IndexType row,

//                    const IndexType* columnIndexes,

//                    const RealType* values,

//                    IndexType elements );



   __cuda_callable__

   void getRowFast( IndexType row,

                    IndexType* columns,

                    RealType* values ) const;

//   __cuda_callable__

//   void getRowFast( IndexType row,

//                    IndexType* columns,

//                    RealType* values ) const;



   __cuda_callable__

   MatrixRow getRow( IndexType row );

src/TNL/Matrices/DistributedMatrix_impl.h

+26 −26
Original line number Diff line number Diff line
@@ -234,32 +234,32 @@ getElementFast( IndexType row, 
   return localMatrix.getElementFast( localRow, column );

}



template< typename Matrix,

          typename Communicator >

__cuda_callable__

bool

DistributedMatrix< Matrix, Communicator >::

setRowFast( IndexType row,

            const IndexType* columnIndexes,

            const RealType* values,

            IndexType elements )

{

   const IndexType localRow = localRowRange.getLocalIndex( row );

   return localMatrix.setRowFast( localRow, columnIndexes, values, elements );

}



template< typename Matrix,

          typename Communicator >

__cuda_callable__

void

DistributedMatrix< Matrix, Communicator >::

getRowFast( IndexType row,

            IndexType* columns,

            RealType* values ) const

{

   const IndexType localRow = localRowRange.getLocalIndex( row );

   return localMatrix.getRowFast( localRow, columns, values );

}

//template< typename Matrix,

//          typename Communicator >

//__cuda_callable__

//bool

//DistributedMatrix< Matrix, Communicator >::

//setRowFast( IndexType row,

//            const IndexType* columnIndexes,

//            const RealType* values,

//            IndexType elements )

//{

//   const IndexType localRow = localRowRange.getLocalIndex( row );

//   return localMatrix.setRowFast( localRow, columnIndexes, values, elements );

//}



//template< typename Matrix,

//          typename Communicator >

//__cuda_callable__

//void

//DistributedMatrix< Matrix, Communicator >::

//getRowFast( IndexType row,

//            IndexType* columns,

//            RealType* values ) const

//{

//   const IndexType localRow = localRowRange.getLocalIndex( row );

//   return localMatrix.getRowFast( localRow, columns, values );

//}



template< typename Matrix,

          typename Communicator >

src/TNL/Matrices/DistributedSpMV.h

+2 −2
Original line number Diff line number Diff line
@@ -85,8 +85,8 @@ public: 
         const auto row = localMatrix->getRow( i );

         bool comm_left = false;

         bool comm_right = false;

         for( IndexType c = 0; c < row.getLength(); c++ ) {

            const IndexType j = row.getElementColumn( c );

         for( IndexType c = 0; c < row.getSize(); c++ ) {

            const IndexType j = row.getColumnIndex( c );

            if( j < columns ) {

               const int owner = Partitioner::getOwner( j, columns, nproc );

               // atomic assignment

src/UnitTests/Matrices/DistributedMatrixTest.h

+5 −5
Original line number Diff line number Diff line
@@ -13,7 +13,7 @@ 
#include <TNL/Communicators/NoDistrCommunicator.h>

#include <TNL/Matrices/DistributedMatrix.h>

#include <TNL/Containers/Partitioner.h>

#include <TNL/Matrices/Legacy/CSR.h>

#include <TNL/Matrices/SparseMatrix.h>



using namespace TNL;
@@ -103,12 +103,12 @@ protected: 


// types for which DistributedMatrixTest is instantiated

using DistributedMatrixTypes = ::testing::Types<

   Matrices::DistributedMatrix< Matrices::Legacy::CSR< double, Devices::Host, int >, Communicators::MpiCommunicator >,

   Matrices::DistributedMatrix< Matrices::Legacy::CSR< double, Devices::Host, int >, Communicators::NoDistrCommunicator >

   Matrices::DistributedMatrix< Matrices::SparseMatrix< double, Devices::Host, int >, Communicators::MpiCommunicator >,

   Matrices::DistributedMatrix< Matrices::SparseMatrix< double, Devices::Host, int >, Communicators::NoDistrCommunicator >

#ifdef HAVE_CUDA

   ,

   Matrices::DistributedMatrix< Matrices::Legacy::CSR< double, Devices::Cuda, int >, Communicators::MpiCommunicator >,

   Matrices::DistributedMatrix< Matrices::Legacy::CSR< double, Devices::Cuda, int >, Communicators::NoDistrCommunicator >

   Matrices::DistributedMatrix< Matrices::SparseMatrix< double, Devices::Cuda, int >, Communicators::MpiCommunicator >,

   Matrices::DistributedMatrix< Matrices::SparseMatrix< double, Devices::Cuda, int >, Communicators::NoDistrCommunicator >

#endif

>;