Commit 3b4ed37c authored by Tomáš Oberhuber's avatar Tomáš Oberhuber
Browse files

Debuging CUDA kernels. 

Clean up makefiles from CXXFLAGS.
parent 04c68239

configure.ac

+1 −1
Original line number Diff line number Diff line
@@ -44,7 +44,7 @@ then 
   STORE_CPPFLAGS=$CPPFLAGS

   STORE_LDFLAGS=$LDFLAGS

   CUDA_LDFLAGS="-L$CUDA_LIBS"

   CUDA_CXXFLAGS="-I$CUDA_HEADERS -DHAVE_CUDA"

   CUDA_CXXFLAGS="-I$CUDA_HEADERS -DHAVE_CUDA -arch sm_13"

   CXXFLAGS="$CXXFLAGS -I$CUDA_HEADERS"

   CPPFLAGS="$CPPFLAGS -I$CUDA_HEADERS"

   LDFLAGS="$LDFLAGS $CUDA_LDFLAGS"

src/Makefile.am

+1 −1
Original line number Diff line number Diff line
@@ -47,7 +47,7 @@ libtnl_mpi_dbg_0_1_la_LIBADD = debug/libtnldebug-mpi-dbg-0.1.la \ 
                       matrix/libtnlmatrix-mpi-dbg-0.1.la

endif                       



tnl_unit_tests_sources = tnl-unit-tests.cpp

tnl_unit_tests_sources = tnl-unit-tests.cu



check_PROGRAMS = tnl-unit-tests

tnl_unit_tests_SOURCES = $(tnl_unit_tests_sources)

src/core/tnlLongVector.h

+1 −0
Original line number Diff line number Diff line
@@ -146,6 +146,7 @@ template< class T > class tnlLongVector : public tnlObject 
   {

      assert( long_vector. GetSize() == GetSize() );

      memcpy( data, long_vector. Data(), GetSize() * sizeof( T ) );

      return true;

   };



#ifdef HAVE_CUDA

src/core/tnlLongVectorCUDA.h

+19 −9
Original line number Diff line number Diff line
@@ -35,10 +35,10 @@ template< class T > class tnlLongVectorCUDA : public tnlObject 
   tnlLongVectorCUDA( int _size = 0 )

    : size( _size ), shared_data( false )

   {

      cudaMalloc( ( void** ) &data, ( size + 1 ) * sizeof( T ) );

      if( ! data )

      if( cudaMalloc( ( void** ) &data, ( size + 1 ) * sizeof( T ) ) != cudaSuccess  )

      {

         cerr << "Unable to allocate new long vector with size " << size << " on CUDA device." << endl;

         data = NULL;

         abort();

      }

      //data ++;
@@ -48,10 +48,10 @@ template< class T > class tnlLongVectorCUDA : public tnlObject 
   tnlLongVectorCUDA( const tnlLongVectorCUDA& v )

    : tnlObject( v ), size( v. size ), shared_data( false )

   {

      cudaMalloc( ( void** ) &data, ( size + 1 ) * sizeof( T ) );

      if( ! data )

      if( cudaMalloc( ( void** ) &data, ( size + 1 ) * sizeof( T ) ) != cudaSuccess )

      {

         cerr << "Unable to allocate new long vector with size " << size << " on CUDA device." << endl;

         data = NULL;

         abort();

      }

      //data ++;
@@ -69,14 +69,14 @@ template< class T > class tnlLongVectorCUDA : public tnlObject 
      if( ! shared_data )

      {

         cudaFree( data );

         data = 0;

         data = NULL;

      }

      size = _size;

      cudaMalloc( ( void** ) &data, size * sizeof( T ) );

      shared_data = false;

      if( ! data )

      if( cudaMalloc( ( void** ) &data, size * sizeof( T ) ) != cudaSuccess )

      {

         cerr << "Unable to allocate new long vector with size " << size << " on CUDA device." << endl;

         data = NULL;

         size = 0;

         return false;

      }
@@ -124,7 +124,12 @@ template< class T > class tnlLongVectorCUDA : public tnlObject 
   bool copyFrom( const tnlLongVector< T >& long_vector )

   {

      assert( long_vector. GetSize() == GetSize() );

      cudaMemcpy( data, long_vector. Data(), GetSize() * sizeof( T ), cudaMemcpyHostToDevice );

      if( cudaMemcpy( data, long_vector. Data(), GetSize() * sizeof( T ), cudaMemcpyHostToDevice ) != cudaSuccess )

      {

         cerr << "Transfer of data from CUDA host ( " << long_vector. GetName()

              << " ) to CUDA device ( " << GetName() << " ) failed." << endl;

         return false;

      }

      return true;

   }
@@ -148,7 +153,12 @@ template< class T > class tnlLongVectorCUDA : public tnlObject 
template< class T > bool tnlLongVector< T > :: copyFrom( const tnlLongVectorCUDA< T >& cuda_vector )

{

   assert( cuda_vector. GetSize() == GetSize() );

   cudaMemcpy( data, cuda_vector. Data(), GetSize() * sizeof( T ), cudaMemcpyDeviceToHost );

   if( cudaMemcpy( data, cuda_vector. Data(), GetSize() * sizeof( T ), cudaMemcpyDeviceToHost ) != cudaSuccess )

   {

      cerr << "Transfer of data from CUDA device ( " << cuda_vector. GetName()

           << " ) to CUDA host ( " << GetName() << " ) failed." << endl;

      return false;

   }

   return true;

}

src/core/tnlLongVectorCUDATester.cu

+1 −1
Original line number Diff line number Diff line
@@ -7,6 +7,6 @@ void testKernelStarter( const float& number, const int size ) 


void testKernelStarter( const double& number, const int size )

{

   testKernel( number, size );

  // testKernel( number, size );

}