Commit 7756e2d0 authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Added Devices::Sequential and corresponding specializations in TNL::Algorithms

parent dbfa5d11

src/TNL/Algorithms/CudaReductionKernel.h

+3 −3
Original line number Diff line number Diff line
@@ -351,7 +351,7 @@ struct CudaReductionKernelLauncher 


      // Copy result on CPU

      Result result;

      MultiDeviceMemoryOperations< Devices::Host, Devices::Cuda >::copy( &result, output, 1 );

      MultiDeviceMemoryOperations< void, Devices::Cuda >::copy( &result, output, 1 );

      return result;

   }
@@ -384,8 +384,8 @@ struct CudaReductionKernelLauncher 
      ////

      // Copy result on CPU

      std::pair< Index, Result > result;

      MultiDeviceMemoryOperations< Devices::Host, Devices::Cuda >::copy( &result.first, idxOutput, 1 );

      MultiDeviceMemoryOperations< Devices::Host, Devices::Cuda >::copy( &result.second, output, 1 );

      MultiDeviceMemoryOperations< void, Devices::Cuda >::copy( &result.first, idxOutput, 1 );

      MultiDeviceMemoryOperations< void, Devices::Cuda >::copy( &result.second, output, 1 );

      return result;

   }

src/TNL/Algorithms/MemoryOperations.h

+3 −3
Original line number Diff line number Diff line
@@ -10,6 +10,7 @@ 


#pragma once



#include <TNL/Devices/Sequential.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

#include <TNL/Cuda/CudaCallable.h>
@@ -17,12 +18,11 @@ 
namespace TNL {

namespace Algorithms {



template< typename DestinationExecution >

template< typename DestinationDevice >

struct MemoryOperations;



// TODO: change "void" to "Execution::Sequential"

template<>

struct MemoryOperations< void >

struct MemoryOperations< Devices::Sequential >

{

   template< typename Element >

   __cuda_callable__

src/TNL/Algorithms/MemoryOperationsHost.hpp

+3 −3
Original line number Diff line number Diff line
@@ -93,7 +93,7 @@ copyFromIterator( DestinationElement* destination, 
                  SourceIterator first,

                  SourceIterator last )

{

   MemoryOperations< void >::copyFromIterator( destination, destinationSize, first, last );

   MemoryOperations< Devices::Sequential >::copyFromIterator( destination, destinationSize, first, last );

}



template< typename DestinationElement,
@@ -137,7 +137,7 @@ containsValue( const Element* data, 
   }

   else {

      // sequential algorithm can return as soon as it finds a match

      return MemoryOperations< void >::containsValue( data, size, value );

      return MemoryOperations< Devices::Sequential >::containsValue( data, size, value );

   }

}
@@ -159,7 +159,7 @@ containsOnlyValue( const Element* data, 
   }

   else {

      // sequential algorithm can return as soon as it finds a mismatch

      return MemoryOperations< void >::containsOnlyValue( data, size, value );

      return MemoryOperations< Devices::Sequential >::containsOnlyValue( data, size, value );

   }

}

src/TNL/Algorithms/MemoryOperationsSequential.hpp

+8 −8
Original line number Diff line number Diff line
@@ -18,7 +18,7 @@ namespace Algorithms { 
template< typename Element >

__cuda_callable__

void

MemoryOperations< void >::

MemoryOperations< Devices::Sequential >::

setElement( Element* data,

            const Element& value )

{
@@ -28,7 +28,7 @@ setElement( Element* data, 
template< typename Element >

__cuda_callable__

Element

MemoryOperations< void >::

MemoryOperations< Devices::Sequential >::

getElement( const Element* data )

{

   return *data;
@@ -37,7 +37,7 @@ getElement( const Element* data ) 
template< typename Element, typename Index >

__cuda_callable__

void

MemoryOperations< void >::

MemoryOperations< Devices::Sequential >::

set( Element* data,

     const Element& value,

     const Index size )
@@ -51,7 +51,7 @@ template< typename DestinationElement, 
          typename Index >

__cuda_callable__

void

MemoryOperations< void >::

MemoryOperations< Devices::Sequential >::

copy( DestinationElement* destination,

      const SourceElement* source,

      const Index size )
@@ -64,7 +64,7 @@ template< typename DestinationElement, 
          typename Index,

          typename SourceIterator >

void

MemoryOperations< void >::

MemoryOperations< Devices::Sequential >::

copyFromIterator( DestinationElement* destination,

                  Index destinationSize,

                  SourceIterator first,
@@ -82,7 +82,7 @@ template< typename Element1, 
          typename Index >

__cuda_callable__

bool

MemoryOperations< void >::

MemoryOperations< Devices::Sequential >::

compare( const Element1* destination,

         const Element2* source,

         const Index size )
@@ -97,7 +97,7 @@ template< typename Element, 
          typename Index >

__cuda_callable__

bool

MemoryOperations< void >::

MemoryOperations< Devices::Sequential >::

containsValue( const Element* data,

               const Index size,

               const Element& value )
@@ -116,7 +116,7 @@ template< typename Element, 
          typename Index >

__cuda_callable__

bool

MemoryOperations< void >::

MemoryOperations< Devices::Sequential >::

containsOnlyValue( const Element* data,

                   const Index size,

                   const Element& value )

src/TNL/Algorithms/Multireduction.h

+30 −0
Original line number Diff line number Diff line
@@ -14,6 +14,7 @@ 


#include <functional>  // reduction functions like std::plus, std::logical_and, std::logical_or etc.



#include <TNL/Devices/Sequential.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>
@@ -23,6 +24,35 @@ namespace Algorithms { 
template< typename Device >

struct Multireduction;



template<>

struct Multireduction< Devices::Sequential >

{

   /**

    * Parameters:

    *    zero: starting value for reduction

    *    dataFetcher: callable object such that `dataFetcher( i, j )` yields

    *                 the i-th value to be reduced from the j-th dataset

    *                 (i = 0,...,size-1; j = 0,...,n-1)

    *    reduction: callable object representing the reduction operation

    *               for example, it can be an instance of std::plus, std::logical_and,

    *               std::logical_or etc.

    *    size: the size of each dataset

    *    n: number of datasets to be reduced

    *    result: output array of size = n

    */

   template< typename Result,

             typename DataFetcher,

             typename Reduction,

             typename Index >

   static constexpr void

   reduce( const Result zero,

           DataFetcher dataFetcher,

           const Reduction reduction,

           const Index size,

           const int n,

           Result* result );

};



template<>

struct Multireduction< Devices::Host >

{