Commit 365229b4 authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Reorganization of benchmarks - moved into src

parent 36639482

CMakeLists.txt

+2 −0
Original line number Diff line number Diff line
@@ -29,6 +29,7 @@ option(WITH_TESTS "Build tests" ON) 
option(WITH_COVERAGE "Enable code coverage reports from unit tests" OFF)

option(WITH_EXAMPLES "Compile the 'examples' directory" ON)

option(WITH_TOOLS "Compile the 'src/Tools' directory" ON)

option(WITH_BENCHMARKS "Compile the 'src/Benchmarks' directory" ON)

option(WITH_PYTHON "Compile the Python bindings" ON)

option(WITH_TEMPLATES_INSTANTIATION "Enable explicit template instantiation" OFF)
@@ -473,6 +474,7 @@ message( " WITH_TESTS=${WITH_TESTS}" ) 
message( "   WITH_COVERAGE=${WITH_COVERAGE}" )

message( "   WITH_EXAMPLES=${WITH_EXAMPLES}" )

message( "   WITH_TOOLS=${WITH_TOOLS}" )

message( "   WITH_BENCHMARKS=${WITH_BENCHMARKS}" )

message( "   WITH_PYTHON=${WITH_PYTHON}" )

message( "   WITH_TEMPLATES_INSTANTIATION=${WITH_TEMPLATES_INSTANTIATION}" )

# Print compiler options

build

+3 −0
Original line number Diff line number Diff line
@@ -26,6 +26,7 @@ WITH_COVERAGE="no" 
WITH_EXAMPLES="yes"

WITH_PYTHON="yes"

WITH_TOOLS="yes"

WITH_BENCHMARKS="yes"



WITH_TEMPLATE_INSTANTIATION="no"

INSTANTIATE_LONG_INT="no"
@@ -60,6 +61,7 @@ do 
        --with-coverage=*                ) WITH_COVERAGE="${option#*=}" ;;

        --with-examples=*                ) WITH_EXAMPLES="${option#*=}" ;;

        --with-tools=*                   ) WITH_TOOLS="${option#*=}" ;;

        --with-benchmarks=*              ) WITH_BENCHMARKS="${option#*=}" ;;

        --with-python=*                  ) WITH_PYTHON="${option#*=}" ;;

        --with-templates-instantiation=* ) WITH_TEMPLATE_INSTANTIATION="${option#*=}" ;;

        --instantiate-long-int=*         ) INSTANTIATE_LONG_INT="${option#*=}" ;;
@@ -164,6 +166,7 @@ cmake_command=( 
         -DWITH_COVERAGE=${WITH_COVERAGE}

         -DWITH_EXAMPLES=${WITH_EXAMPLES}

         -DWITH_TOOLS=${WITH_TOOLS}

         -DWITH_BENCHMARKS=${WITH_BENCHMARKS}

         -DWITH_PYTHON=${WITH_PYTHON}

         -DDCMTK_DIR=${DCMTK_DIR}

         -DWITH_TEMPLATE_INSTANTIATION=${WITH_TEMPLATE_INSTANTIATION}

src/Benchmarks/BLAS/CMakeLists.txt

0 → 100644
+10 −0
Original line number Diff line number Diff line 
if( BUILD_CUDA )

    CUDA_ADD_EXECUTABLE( tnl-benchmark-blas tnl-benchmark-blas.cu )

    CUDA_ADD_CUBLAS_TO_TARGET( tnl-benchmark-blas )

    TARGET_LINK_LIBRARIES( tnl-benchmark-blas tnl )

else()

    ADD_EXECUTABLE( tnl-benchmark-blas tnl-benchmark-blas.cpp )

    TARGET_LINK_LIBRARIES( tnl-benchmark-blas tnl )

endif()



install( TARGETS tnl-benchmark-blas RUNTIME DESTINATION bin )

src/Benchmarks/BLAS/array-operations.h

0 → 100644
+165 −0
Original line number Diff line number Diff line 
/***************************************************************************

                          array-operations.h  -  description

                             -------------------

    begin                : Dec 30, 2015

    copyright            : (C) 2015 by Tomas Oberhuber et al.

    email                : tomas.oberhuber@fjfi.cvut.cz

 ***************************************************************************/



/* See Copyright Notice in tnl/Copyright */



// Implemented by: Jakub Klinkovsky



#pragma once



#include "../Benchmarks.h"



#include <TNL/Containers/Array.h>



namespace TNL {

namespace Benchmarks {



template< typename Real = double,

          typename Index = int >

bool

benchmarkArrayOperations( Benchmark & benchmark,

                          const int & loops,

                          const long & size )

{

   typedef Containers::Array< Real, Devices::Host, Index > HostArray;

   typedef Containers::Array< Real, Devices::Cuda, Index > CudaArray;

   using namespace std;



   double datasetSize = ( double ) ( loops * size ) * sizeof( Real ) / oneGB;



   HostArray hostArray, hostArray2;

   CudaArray deviceArray, deviceArray2;

   hostArray.setSize( size );

   hostArray2.setSize( size );

#ifdef HAVE_CUDA

   deviceArray.setSize( size );

   deviceArray2.setSize( size );

#endif



   Real resultHost, resultDevice;





   // reset functions

   auto reset1 = [&]() {

      hostArray.setValue( 1.0 );

#ifdef HAVE_CUDA

      deviceArray.setValue( 1.0 );

#endif

   };

   auto reset2 = [&]() {

      hostArray2.setValue( 1.0 );

#ifdef HAVE_CUDA

      deviceArray2.setValue( 1.0 );

#endif

   };

   auto reset12 = [&]() {

      reset1();

      reset2();

   };





   reset12();





   auto compareHost = [&]() {

      resultHost = (int) hostArray == hostArray2;

   };

   auto compareCuda = [&]() {

      resultDevice = (int) deviceArray == deviceArray2;

   };

   benchmark.setOperation( "comparison (operator==)", 2 * datasetSize );

   benchmark.time( reset1, "CPU", compareHost );

#ifdef HAVE_CUDA

   benchmark.time( reset1, "GPU", compareCuda );

#endif





   auto copyAssignHostHost = [&]() {

      hostArray = hostArray2;

   };

   auto copyAssignCudaCuda = [&]() {

      deviceArray = deviceArray2;

   };

   benchmark.setOperation( "copy (operator=)", 2 * datasetSize );

   // copyBasetime is used later inside HAVE_CUDA guard, so the compiler will

   // complain when compiling without CUDA

   const double copyBasetime = benchmark.time( reset1, "CPU", copyAssignHostHost );

#ifdef HAVE_CUDA

   benchmark.time( reset1, "GPU", copyAssignCudaCuda );

#endif





   auto copyAssignHostCuda = [&]() {

      deviceArray = hostArray;

   };

   auto copyAssignCudaHost = [&]() {

      hostArray = deviceArray;

   };

#ifdef HAVE_CUDA

   benchmark.setOperation( "copy (operator=)", datasetSize, copyBasetime );

   benchmark.time( reset1,

                   "CPU->GPU", copyAssignHostCuda,

                   "GPU->CPU", copyAssignCudaHost );

#endif





   auto setValueHost = [&]() {

      hostArray.setValue( 3.0 );

   };

   auto setValueCuda = [&]() {

      deviceArray.setValue( 3.0 );

   };

   benchmark.setOperation( "setValue", datasetSize );

   benchmark.time( reset1, "CPU", setValueHost );

#ifdef HAVE_CUDA

   benchmark.time( reset1, "GPU", setValueCuda );

#endif





   auto setSizeHost = [&]() {

      hostArray.setSize( size );

   };

   auto setSizeCuda = [&]() {

      deviceArray.setSize( size );

   };

   auto resetSize1 = [&]() {

      hostArray.reset();

#ifdef HAVE_CUDA

      deviceArray.reset();

#endif

   };

   benchmark.setOperation( "allocation (setSize)", datasetSize );

   benchmark.time( resetSize1, "CPU", setSizeHost );

#ifdef HAVE_CUDA

   benchmark.time( resetSize1, "GPU", setSizeCuda );

#endif





   auto resetSizeHost = [&]() {

      hostArray.reset();

   };

   auto resetSizeCuda = [&]() {

      deviceArray.reset();

   };

   auto setSize1 = [&]() {

      hostArray.setSize( size );

#ifdef HAVE_CUDA

      deviceArray.setSize( size );

#endif

   };

   benchmark.setOperation( "deallocation (reset)", datasetSize );

   benchmark.time( setSize1, "CPU", resetSizeHost );

#ifdef HAVE_CUDA

   benchmark.time( setSize1, "GPU", resetSizeCuda );

#endif



   return true;

}



} // namespace Benchmarks

} // namespace TNL

tests/benchmarks/cublasWrappers.h→src/Benchmarks/BLAS/cublasWrappers.h

+14 −14

File changed and moved.

Contains only whitespace changes.