Commit 81232c66 authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Moved CudaMultireductionKernel, CudaReductionKernel and CudaScanKernel into a 'detail' namespace 

The kernels are not interesting for users and Doxygen made a mess out of
various CUDA declarations. The 'detail' namespace is considered internal
and excluded from processing by Doxygen.
parent 3c7d5d42

src/TNL/Algorithms/Multireduction.hpp

+2 −2
Original line number Diff line number Diff line
@@ -19,7 +19,7 @@ 
#include <TNL/Assert.h>

#include <TNL/Algorithms/Multireduction.h>

#include <TNL/Algorithms/MultiDeviceMemoryOperations.h>

#include <TNL/Algorithms/CudaMultireductionKernel.h>

#include <TNL/Algorithms/detail/CudaMultireductionKernel.h>



#ifdef CUDA_REDUCTION_PROFILING

#include <TNL/Timer.h>
@@ -212,7 +212,7 @@ reduce( const Result zero, 


   // start the reduction on the GPU

   Result* deviceAux1 = nullptr;

   const int reducedSize = CudaMultireductionKernelLauncher( zero, dataFetcher, reduction, size, n, deviceAux1 );

   const int reducedSize = detail::CudaMultireductionKernelLauncher( zero, dataFetcher, reduction, size, n, deviceAux1 );



   #ifdef CUDA_REDUCTION_PROFILING

      timer.stop();

src/TNL/Algorithms/Reduction.hpp

+3 −3
Original line number Diff line number Diff line
@@ -17,7 +17,7 @@ 
//#define CUDA_REDUCTION_PROFILING



#include <TNL/Algorithms/Reduction.h>

#include <TNL/Algorithms/CudaReductionKernel.h>

#include <TNL/Algorithms/detail/CudaReductionKernel.h>

#include <TNL/Algorithms/MultiDeviceMemoryOperations.h>



#ifdef CUDA_REDUCTION_PROFILING
@@ -311,7 +311,7 @@ reduce( const Index begin, 
      timer.start();

   #endif



   CudaReductionKernelLauncher< Index, Result > reductionLauncher( begin, end );

   detail::CudaReductionKernelLauncher< Index, Result > reductionLauncher( begin, end );



   // start the reduce on the GPU

   Result* deviceAux1( 0 );
@@ -401,7 +401,7 @@ reduceWithArgument( const Index begin, 
      timer.start();

   #endif



   CudaReductionKernelLauncher< Index, Result > reductionLauncher( begin, end );

   detail::CudaReductionKernelLauncher< Index, Result > reductionLauncher( begin, end );



   // start the reduce on the GPU

   Result* deviceAux1( nullptr );

src/TNL/Algorithms/Scan.hpp

+4 −4
Original line number Diff line number Diff line
@@ -17,7 +17,7 @@ 
#include <TNL/Assert.h>

#include <TNL/Containers/Array.h>

#include <TNL/Containers/StaticArray.h>

#include <TNL/Algorithms/CudaScanKernel.h>

#include <TNL/Algorithms/detail/CudaScanKernel.h>

#include <TNL/Exceptions/CudaSupportMissing.h>

#include <TNL/Exceptions/NotImplementedError.h>
@@ -227,7 +227,7 @@ perform( Vector& v, 
   using RealType = typename Vector::RealType;

   using IndexType = typename Vector::IndexType;



   CudaScanKernelLauncher< Type, RealType, IndexType >::perform(

   detail::CudaScanKernelLauncher< Type, RealType, IndexType >::perform(

      end - begin,

      &v.getData()[ begin ],  // input

      &v.getData()[ begin ],  // output
@@ -253,7 +253,7 @@ performFirstPhase( Vector& v, 
   using RealType = typename Vector::RealType;

   using IndexType = typename Vector::IndexType;



   return CudaScanKernelLauncher< Type, RealType, IndexType >::performFirstPhase(

   return detail::CudaScanKernelLauncher< Type, RealType, IndexType >::performFirstPhase(

      end - begin,

      &v.getData()[ begin ],  // input

      &v.getData()[ begin ],  // output
@@ -281,7 +281,7 @@ performSecondPhase( Vector& v, 
   using RealType = typename Vector::RealType;

   using IndexType = typename Vector::IndexType;



   CudaScanKernelLauncher< Type, RealType, IndexType >::performSecondPhase(

   detail::CudaScanKernelLauncher< Type, RealType, IndexType >::performSecondPhase(

      end - begin,

      &v.getData()[ begin ],  // output

      blockShifts.getData(),

src/TNL/Algorithms/CudaMultireductionKernel.h→src/TNL/Algorithms/detail/CudaMultireductionKernel.h

+2 −0
Original line number Diff line number Diff line
@@ -21,6 +21,7 @@ 


namespace TNL {

namespace Algorithms {

namespace detail {



#ifdef HAVE_CUDA

/****
@@ -281,5 +282,6 @@ CudaMultireductionKernelLauncher( const Result zero, 
#endif

}



} // namespace detail

} // namespace Algorithms

} // namespace TNL

src/TNL/Algorithms/CudaReductionKernel.h→src/TNL/Algorithms/detail/CudaReductionKernel.h

+2 −0
Original line number Diff line number Diff line
@@ -21,6 +21,7 @@ 


namespace TNL {

namespace Algorithms {

namespace detail {



/****

 * The performance of this kernel is very sensitive to register usage.
@@ -642,5 +643,6 @@ struct CudaReductionKernelLauncher 
      Index reducedSize;

};



} // namespace detail

} // namespace Algorithms

} // namespace TNL