Commit c4cc606a authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Fixed reduction of segments to work even for empty matrices

parent ce8f92fc

src/TNL/Algorithms/Segments/EllpackView.hpp

+4 −0
Original line number Diff line number Diff line
@@ -105,6 +105,8 @@ struct EllpackCudaReductionDispatcher 
   exec( Index first, Index last, Fetch& fetch, const Reduction& reduction, ResultKeeper& keeper, const Real& zero, Index segmentSize )

   {

   #ifdef HAVE_CUDA

      if( last <= first )

         return;

      const Index segmentsCount = last - first;

      const Index threadsCount = segmentsCount * 32;

      const Index blocksCount = Cuda::getNumberOfBlocks( threadsCount, 256 );
@@ -128,6 +130,8 @@ struct EllpackCudaReductionDispatcher< Index, Fetch, Reduction, ResultKeeper, Re 
   exec( Index first, Index last, Fetch& fetch, const Reduction& reduction, ResultKeeper& keeper, const Real& zero, Index segmentSize )

   {

   #ifdef HAVE_CUDA

      if( last <= first )

         return;

      const Index segmentsCount = last - first;

      const Index threadsCount = segmentsCount * 32;

      const Index blocksCount = Cuda::getNumberOfBlocks( threadsCount, 256 );

src/TNL/Algorithms/Segments/Kernels/CSRHybridKernel.hpp

+2 −0
Original line number Diff line number Diff line
@@ -245,6 +245,8 @@ reduceSegments( const OffsetsView& offsets, 
    TNL_ASSERT_LE( this->threadsPerSegment, ThreadsInBlock, "" );



#ifdef HAVE_CUDA

    if( last <= first )

       return;

    const size_t threadsCount = this->threadsPerSegment * ( last - first );

    dim3 blocksCount, gridsCount, blockSize( ThreadsInBlock );

    TNL::Cuda::setupThreads( blockSize, blocksCount, gridsCount, threadsCount );

src/TNL/Algorithms/Segments/Kernels/CSRLightKernel.hpp

+3 −0
Original line number Diff line number Diff line
@@ -424,6 +424,9 @@ struct CSRLightKernelreduceSegmentsDispatcher< Index, Device, Fetch, Reduce, Kee 
                       const Index threadsPerSegment )

   {

#ifdef HAVE_CUDA

    if( last <= first )

       return;



      const size_t threads = 128;

      Index blocks, groupSize;

src/TNL/Algorithms/Segments/Kernels/CSRVectorKernel.hpp

+3 −0
Original line number Diff line number Diff line
@@ -140,6 +140,9 @@ reduceSegments( const OffsetsView& offsets, 
                         Args... args )

{

#ifdef HAVE_CUDA

    if( last <= first )

       return;



    const Index warpsCount = last - first;

    const size_t threadsCount = warpsCount * TNL::Cuda::getWarpSize();

    dim3 blocksCount, gridsCount, blockSize( 256 );

src/UnitTests/Matrices/SparseMatrixTest.hpp

+3 −0
Original line number Diff line number Diff line
@@ -243,6 +243,9 @@ void test_SetDimensions() 


   EXPECT_EQ( m.getRows(), 9 );

   EXPECT_EQ( m.getColumns(), 8 );



   // test empty matrix

   m.setDimensions( 0, 0 );

}



template< typename Matrix >