Commit 1fe62640 authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Replaced static member variables in CudaPrefixSumKernelLauncher with static getters

parent af6d1d6b

src/TNL/Containers/Algorithms/CudaPrefixSumKernel.h

+17 −24
Original line number Diff line number Diff line
@@ -297,8 +297,7 @@ struct CudaPrefixSumKernelLauncher 
       */

      const Index elementsInBlock = 8 * blockSize;

      const Index numberOfBlocks = roundUpDivision( size, elementsInBlock );

      //const auto maxGridSize = 3; //Devices::Cuda::getMaxGridSize();

      const Index numberOfGrids = Devices::Cuda::getNumberOfGrids( numberOfBlocks, maxGridSize );

      const Index numberOfGrids = Devices::Cuda::getNumberOfGrids( numberOfBlocks, maxGridSize() );

      Real gridShift = zero;

      //std::cerr << "numberOfgrids =  " << numberOfGrids << std::endl;
@@ -310,10 +309,10 @@ struct CudaPrefixSumKernelLauncher 
         /****

          * Compute current grid size and size of data to be scanned

          */

         const Index gridOffset = gridIdx * maxGridSize * elementsInBlock;

         const Index gridOffset = gridIdx * maxGridSize() * elementsInBlock;

         Index currentSize = size - gridOffset;

         if( currentSize / elementsInBlock > maxGridSize )

            currentSize = maxGridSize * elementsInBlock;

         if( currentSize / elementsInBlock > maxGridSize() )

            currentSize = maxGridSize() * elementsInBlock;



         //std::cerr << "GridIdx = " << gridIdx << " grid size = " << currentSize << std::endl;

         cudaRecursivePrefixSum( prefixSumType,
@@ -331,37 +330,31 @@ struct CudaPrefixSumKernelLauncher 
       * Store the number of CUDA grids for the purpose of unit testing, i.e.

       * to check if we test the algorithm with more than one CUDA grid.

       */

      gridsCount = numberOfGrids;

      gridsCount() = numberOfGrids;

   }



   /****

    * The following serves for setting smaller maxGridSize so that we can force

    * the prefix sum in CUDA to run with more the one grids in unit tests.

    */

   static void setMaxGridSize( int newMaxGridSize ) {

      maxGridSize = newMaxGridSize;

   static int& maxGridSize()

   {

      static int maxGridSize = Devices::Cuda::getMaxGridSize();

      return maxGridSize;

   }



   static void resetMaxGridSize() {

      maxGridSize = Devices::Cuda::getMaxGridSize();

   static void resetMaxGridSize()

   {

      maxGridSize() = Devices::Cuda::getMaxGridSize();

   }



   static int maxGridSize;



   static int gridsCount;

   static int& gridsCount()

   {

      static int gridsCount = -1;

      return gridsCount;

   }

};



template< PrefixSumType prefixSumType,

          typename Real,

          typename Index >

int CudaPrefixSumKernelLauncher< prefixSumType, Real, Index >::maxGridSize = Devices::Cuda::getMaxGridSize();



template< PrefixSumType prefixSumType,

          typename Real,

          typename Index >

int CudaPrefixSumKernelLauncher< prefixSumType, Real, Index >::gridsCount = -1;





#endif



} // namespace Algorithms

src/TNL/Devices/Host.h

+7 −5
Original line number Diff line number Diff line
@@ -113,17 +113,19 @@ public: 
   }



   protected:

      static bool& ompEnabled() {

      static bool& ompEnabled()

      {

#ifdef HAVE_OPENMP

         static bool ompEnabled( true );

         static bool ompEnabled = true;

#else

         static bool ompEnabled( false );

         static bool ompEnabled = false;

#endif

         return ompEnabled;

      }



      static int& maxThreadsCount() {

         static int maxThreadsCount( -1 );

      static int& maxThreadsCount()

      {

         static int maxThreadsCount = -1;

         return maxThreadsCount;

      }

};

src/UnitTests/Containers/VectorPrefixSumTest.h

+13 −13
Original line number Diff line number Diff line
@@ -74,11 +74,11 @@ TYPED_TEST( VectorTest, prefixSum ) 
   if( std::is_same< DeviceType, Devices::Cuda >::value )

   {

#ifdef HAVE_CUDA

      Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::setMaxGridSize( 3 );

      Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::maxGridSize() = 3;

      v = 0;

      v_host = -1;

      v.prefixSum();

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::gridsCount ), 1  );

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::gridsCount() ), 1  );

      v_host = v;

      for( int i = 0; i < size; i++ )

         EXPECT_EQ( v_host[ i ], 0 );
@@ -86,7 +86,7 @@ TYPED_TEST( VectorTest, prefixSum ) 
      setLinearSequence( v );

      v_host = -1;

      v.prefixSum();

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::gridsCount ), 1  );

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::gridsCount() ), 1  );

      v_host = v;

      for( int i = 1; i < size; i++ )

         EXPECT_EQ( v_host[ i ] - v_host[ i - 1 ], i );
@@ -94,7 +94,7 @@ TYPED_TEST( VectorTest, prefixSum ) 
      setConstantSequence( v, 1 );

      v_host = -1;

      v_view.prefixSum();

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::gridsCount ), 1  );

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::gridsCount() ), 1  );

      v_host = v_view;

      for( int i = 0; i < size; i++ )

         EXPECT_EQ( v_host[ i ], i + 1 );
@@ -102,7 +102,7 @@ TYPED_TEST( VectorTest, prefixSum ) 
      v = 0;

      v_host = -1;

      v_view.prefixSum();

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::gridsCount ), 1  );

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::gridsCount() ), 1  );

      v_host = v_view;

      for( int i = 0; i < size; i++ )

         EXPECT_EQ( v_host[ i ], 0 );
@@ -110,7 +110,7 @@ TYPED_TEST( VectorTest, prefixSum ) 
      setLinearSequence( v );

      v_host = -1;

      v_view.prefixSum();

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::gridsCount ), 1  );

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Inclusive, RealType, IndexType >::gridsCount() ), 1  );

      v_host = v_view;

      for( int i = 1; i < size; i++ )

         EXPECT_EQ( v_host[ i ] - v_host[ i - 1 ], i );
@@ -184,12 +184,12 @@ TYPED_TEST( VectorTest, exclusivePrefixSum ) 
   if( std::is_same< DeviceType, Devices::Cuda >::value )

   {

#ifdef HAVE_CUDA

      Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::setMaxGridSize( 3 );

      Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::maxGridSize() = 3;



      setConstantSequence( v, 1 );

      v_host = -1;

      v.template prefixSum< Algorithms::PrefixSumType::Exclusive >();

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount ), 1 );

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount() ), 1 );

      v_host = v;

      for( int i = 0; i < size; i++ )

         EXPECT_EQ( v_host[ i ], i );
@@ -197,7 +197,7 @@ TYPED_TEST( VectorTest, exclusivePrefixSum ) 
      v.setValue( 0 );

      v_host = -1;

      v.template prefixSum< Algorithms::PrefixSumType::Exclusive >();

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount ), 1 );

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount() ), 1 );

      v_host = v;

      for( int i = 0; i < size; i++ )

         EXPECT_EQ( v_host[ i ], 0 );
@@ -205,7 +205,7 @@ TYPED_TEST( VectorTest, exclusivePrefixSum ) 
      setLinearSequence( v );

      v_host = -1;

      v.template prefixSum< Algorithms::PrefixSumType::Exclusive >();

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount ), 1 );

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount() ), 1 );

      v_host = v;

      for( int i = 1; i < size; i++ )

         EXPECT_EQ( v_host[ i ] - v_host[ i - 1 ], i - 1 );
@@ -213,7 +213,7 @@ TYPED_TEST( VectorTest, exclusivePrefixSum ) 
      setConstantSequence( v, 1 );

      v_host = -1;

      v_view.template prefixSum< Algorithms::PrefixSumType::Exclusive >();

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount ), 1 );

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount() ), 1 );

      v_host = v_view;

      for( int i = 0; i < size; i++ )

         EXPECT_EQ( v_host[ i ], i );
@@ -221,7 +221,7 @@ TYPED_TEST( VectorTest, exclusivePrefixSum ) 
      v.setValue( 0 );

      v_host = -1;

      v_view.template prefixSum< Algorithms::PrefixSumType::Exclusive >();

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount ), 1 );

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount() ), 1 );

      v_host = v_view;

      for( int i = 0; i < size; i++ )

         EXPECT_EQ( v_host[ i ], 0 );
@@ -229,7 +229,7 @@ TYPED_TEST( VectorTest, exclusivePrefixSum ) 
      setLinearSequence( v );

      v_host = -1;

      v_view.template prefixSum< Algorithms::PrefixSumType::Exclusive >();

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount ), 1 );

      EXPECT_GT( ( Algorithms::CudaPrefixSumKernelLauncher< Algorithms::PrefixSumType::Exclusive, RealType, IndexType >::gridsCount() ), 1 );

      v_host = v_view;

      for( int i = 1; i < size; i++ )

         EXPECT_EQ( v_host[ i ] - v_host[ i - 1 ], i - 1 );