Commit e2c5a3db authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Removed CUBLAS because it is used only in the benchmarks 

And of course our implementation is faster ;-)
parent 052e268e

CMakeLists.txt

+0 −17
Original line number Diff line number Diff line
@@ -161,23 +161,6 @@ if( WITH_CUDA STREQUAL "yes" ) 
        # TODO: this is necessary only due to a bug in cmake

        set( CUDA_ADD_LIBRARY_OPTIONS -shared )



        ####

        # Check for cuBLAS

        #

        if( NOT WITH_CUBLAS STREQUAL "no" )

            find_path( CUBLAS_INCLUDE_DIR cublas_v2.h

                       /usr/local/cuda/include

                       ${CUDA_INCLUDE_DIR}

                       DOC "CUBLAS headers." )

            if( ${CUBLAS_INCLUDE_DIR} STREQUAL "CUBLAS_INCLUDE_DIR-NOTFOUND" )

                message( "CUBLAS not found." )

                set( HAVE_CUBLAS "//#define HAVE_CUBLAS 1" )

            else()

                message( "CUBLAS found. -- ${CUBLAS_INCLUDE_DIR}" )

                set( HAVE_CUBLAS "#define HAVE_CUBLAS 1" )

            endif()

        endif( NOT WITH_CUBLAS STREQUAL "no" )



        ####

        # Check for CUSP

        #

src/TNL/CMakeLists.txt

+0 −6
Original line number Diff line number Diff line
@@ -87,9 +87,6 @@ set( tnl_CUDA__SOURCES ${tnl_config_CUDA__SOURCES} 
if( BUILD_CUDA )

   CUDA_ADD_LIBRARY( tnl SHARED ${tnl_CUDA__SOURCES}

                     OPTIONS ${CUDA_ADD_LIBRARY_OPTIONS} )

   if( HAVE_CUBLAS )

      CUDA_ADD_CUBLAS_TO_TARGET( tnl )

   endif( HAVE_CUBLAS )

   CUDA_ADD_LIBRARY( tnl_static STATIC ${tnl_CUDA__SOURCES} )

else( BUILD_CUDA )

   ADD_LIBRARY( tnl SHARED ${tnl_SOURCES} )
@@ -111,9 +108,6 @@ IF( BUILD_MPI ) 
   if( BUILD_CUDA )

      CUDA_ADD_LIBRARY( tnl-mpi SHARED ${tnl_CUDA__SOURCES}

                        OPTIONS ${CUDA_ADD_LIBRARY_OPTIONS} )

      if( HAVE_CUBLAS )

         CUDA_ADD_CUBLAS_TO_TARGET( tnl-mpi )

      endif( HAVE_CUBLAS )

      CUDA_ADD_LIBRARY( tnl-mpi_static STATIC ${tnl_CUDA__SOURCES} )

   else( BUILD_CUDA )

      ADD_LIBRARY( tnl-mpi SHARED ${tnl_SOURCES} )

src/TNL/Containers/Algorithms/CMakeLists.txt

+0 −1
Original line number Diff line number Diff line
@@ -6,7 +6,6 @@ set( headers ArrayOperations.h 
             ArrayOperationsMIC_impl.h

             cuda-prefix-sum.h

             cuda-prefix-sum_impl.h

             CublasWrapper.h

             CudaMultireductionKernel.h

             CudaReductionBuffer.h

             CudaReductionKernel.h

src/TNL/Containers/Algorithms/CublasWrapper.h

deleted100644 → 0
+0 −68
Original line number Diff line number Diff line 
/***************************************************************************

                          CublasWraper.h  -  description

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

    begin                : Apr 7, 2015

    copyright            : (C) 2015 by Tomas Oberhuber

    email                : tomas.oberhuber@fjfi.cvut.cz

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



/* See Copyright Notice in tnl/Copyright */



#pragma once



#if defined HAVE_CUBLAS && defined HAVE_CUDA

#include <cublas_v2.h>

#endif



namespace TNL {

namespace Containers {

namespace Algorithms {



template< typename Real1,

          typename Real2,

          typename Index >

class CublasWraper

{

    public:

        static bool dot( const Real1* v1, const Real2* v2, const Index size, Real1& result)

        {

            return false;

        }

};



#if defined HAVE_CUBLAS && defined HAVE_CUDA



template< typename Index >

class CublasWraper< float, float, Index >

{

    public:

        static bool dot( const float* v1, const float* v2, const Index size, float& result)

        {



            cublasHandle_t handle;

            cublasCreate( &handle );

            cublasSdot( handle, size, v1, 1, v2, 1, &result );

            cublasDestroy( handle );

            return false;

        }

};



template< typename Index >

class CublasWraper< double, double, Index >

{

    public:

        static bool dot( const double* v1, const double* v2, const Index size, double& result)

        {

            cublasHandle_t handle;

            cublasCreate( &handle );

            cublasDdot( handle, size, v1, 1, v2, 1, &result );

            cublasDestroy( handle );

            return false;

        }

};

#endif



} // namespace Algorithms

} // namespace Containers

} // namespace TNL

src/TNL/Containers/Algorithms/VectorOperationsCuda_impl.h

+0 −7
Original line number Diff line number Diff line
@@ -14,7 +14,6 @@ 
#include <TNL/Exceptions/CudaSupportMissing.h>

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

#include <TNL/Containers/Algorithms/cuda-prefix-sum.h>

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



namespace TNL {

namespace Containers {
@@ -426,12 +425,6 @@ getScalarProduct( const Vector1& v1, 
   TNL_ASSERT_EQ( v1.getSize(), v2.getSize(), "The vector sizes must be the same." );



   Real result( 0 );

/*#if defined HAVE_CUBLAS && defined HAVE_CUDA

   if( CublasWraper< typename Vector1::RealType,

                         typename Vector2::RealType,

                         typename Vector1::IndexType >::dot( v1.getData(), v1.getData(), v1.getSize(), result ) )

       return result;

#endif*/

   Algorithms::ParallelReductionScalarProduct< typename Vector1::RealType, typename Vector2::RealType > operation;

   Reduction< Devices::Cuda >::reduce( operation,

                                       v1.getSize(),