Commit 35f8fd27 authored by Tomáš Oberhuber's avatar Tomáš Oberhuber
Browse files

Parallel reduction tutorial.

parent d4d0a13e

Documentation/Tutorials/Arrays/tutorial_01_Arrays.md

+3 −0
Original line number Diff line number Diff line
@@ -15,6 +15,7 @@ This tutorial introduces arrays in TNL. Array is one of the most important struc 
   5. [Checking the array contents](#checking_the_array_contents)

   6. [IO operations with arrays](#io_operations_with-arrays)

2. [Static arrays](#static_arrays)

2. [Distributed arrays](#distributed_arrays)



## Arrays <a name="arrays"></a>
@@ -148,3 +149,5 @@ Output: 
\include ArrayIO.out



## Static arrays <a name="static_arrays"></a>



## Distributed arrays <a name="distributed_arrays"></a>

Documentation/Tutorials/Reduction/CMakeLists.txt

+12 −3
Original line number Diff line number Diff line 
IF( BUILD_CUDA )

#   CUDA_ADD_EXECUTABLE( ArrayAllocation ArrayAllocation.cu )

#   ADD_CUSTOM_COMMAND( COMMAND ArrayAllocation > ArrayAllocation.out OUTPUT ArrayAllocation.out )

   CUDA_ADD_EXECUTABLE( SumExample SumExample.cu )

   ADD_CUSTOM_COMMAND( COMMAND SumExample > SumExample.out OUTPUT SumExample.out )

   CUDA_ADD_EXECUTABLE( ProductExample ProductExample.cu )

   ADD_CUSTOM_COMMAND( COMMAND ProductExample > ProductExample.out OUTPUT ProductExample.out )

   CUDA_ADD_EXECUTABLE( ScalarProductExample ScalarProductExample.cu )

   ADD_CUSTOM_COMMAND( COMMAND ScalarProductExample > ScalarProductExample.out OUTPUT ScalarProductExample.out )

   CUDA_ADD_EXECUTABLE( MaximumNormExample MaximumNormExample.cu )

   ADD_CUSTOM_COMMAND( COMMAND MaximumNormExample > MaximumNormExample.out OUTPUT MaximumNormExample.out )

ENDIF()



IF( BUILD_CUDA )

ADD_CUSTOM_TARGET( TutorialsReduction-cuda ALL DEPENDS

 )

   SumExample.out

   ProductExample.out

   ScalarProductExample.out

   MaximumNormExample.out )

ENDIF()

Documentation/Tutorials/Reduction/MaximumNormExample.cpp

0 → 100644
+34 −0
Original line number Diff line number Diff line 
#include <iostream>

#include <cstdlib>

#include <TNL/Containers/Vector.h>

#include <TNL/Containers/Algorithms/Reduction.h>



using namespace TNL;

using namespace TNL::Containers;

using namespace TNL::Containers::Algorithms;



template< typename Device >

double maximumNorm( const Vector< double, Device >& v )

{

   auto view = v.getView();

   auto fetch = [=] __cuda_callable__ ( int i ) { return abs( view[ i ] ); };

   auto reduce = [] __cuda_callable__ ( double& a, const double& b ) { a = max( a, b ); };

   auto volatileReduce = [=] __cuda_callable__ ( volatile double& a, const volatile double& b ) { a = max( a ,b ); };

   return Reduction< Device >::reduce( v.getSize(), reduce, volatileReduce, fetch, 0.0 );

}



int main( int argc, char* argv[] )

{

   Vector< double, Devices::Host > host_v( 10 );

   host_v.evaluate( [] __cuda_callable__ ( int i )->double { return i - 7; } );

   std::cout << "host_v = " << host_v << std::cout;

   std::cout << "The maximum norm of the host vector elements is " << maximumNorm( host_v ) << "." << std::endl;

#ifdef HAVE_CUDA

   Vector< double, Devices::Cuda > cuda_v( 10 );

   cuda_v.evaluate( [] __cuda_callable__ ( int i )->double { return i - 7; } );

   std::cout << "cuda_v = " << cuda_v << std::cout;

   std::cout << "The maximum norm of the CUDA vector elements is " << maximumNorm( cuda_v ) << "." << std::endl;

#endif

   return EXIT_SUCCESS;

}

Documentation/Tutorials/Reduction/MaximumNormExample.cu

0 → 120000
+1 −0
Original line number Diff line number Diff line 
MaximumNormExample.cpp
 
 No newline at end of file

Documentation/Tutorials/Reduction/ProductExample.cpp

0 → 100644
+34 −0
Original line number Diff line number Diff line 
#include <iostream>

#include <cstdlib>

#include <TNL/Containers/Vector.h>

#include <TNL/Containers/Algorithms/Reduction.h>



using namespace TNL;

using namespace TNL::Containers;

using namespace TNL::Containers::Algorithms;



template< typename Device >

double product( const Vector< double, Device >& v )

{

   auto view = v.getView();

   auto fetch = [=] __cuda_callable__ ( int i ) { return view[ i ]; };

   auto reduce = [] __cuda_callable__ ( double& a, const double& b ) { a *= b; };

   auto volatileReduce = [=] __cuda_callable__ ( volatile double& a, const volatile double& b ) { a *= b; };

   return Reduction< Device >::reduce( v.getSize(), reduce, volatileReduce, fetch, 1.0 );

}



int main( int argc, char* argv[] )

{

   Vector< double, Devices::Host > host_v( 10 );

   host_v = 1.0;

   std::cout << "host_v = " << host_v << std::cout;

   std::cout << "The product of the host vector elements is " << product( host_v ) << "." << std::endl;

#ifdef HAVE_CUDA

   Vector< double, Devices::Cuda > cuda_v( 10 );

   cuda_v = 1.0;

   std::cout << "cuda_v = " << cuda_v << std::cout;

   std::cout << "The product of the CUDA vector elements is " << product( cuda_v ) << "." << std::endl;

#endif

   return EXIT_SUCCESS;

}