FLA_Copy_external_gpu.c File Reference

(r12600)

Functions
FLA_Error	FLA_Copy_external_gpu (FLA_Obj A, void *A_gpu, FLA_Obj B, void *B_gpu)

---

Function Documentation

FLA_Error FLA_Copy_external_gpu	(	FLA_Obj	A,
		void *	A_gpu,
		FLA_Obj	B,
		void *	B_gpu
	)

References FLA_Check_error_level(), FLA_Copy_check(), FLA_Obj_datatype(), FLA_Obj_has_zero_dim(), FLA_Obj_length(), and FLA_Obj_width().

Referenced by FLASH_Queue_exec_task_gpu().

{
  FLA_Datatype datatype;
  int          m_B, n_B;
  int          ldim_A, inc_A;
  int          ldim_B, inc_B;
  int          i;

  if ( FLA_Check_error_level() == FLA_FULL_ERROR_CHECKING ) 
    FLA_Copy_check( A, B );

  if ( FLA_Obj_has_zero_dim( A ) ) return FLA_SUCCESS;

  // It is important that we get the datatype of B and not A, since A could
  // be an FLA_CONSTANT.
  datatype = FLA_Obj_datatype( B );

  ldim_A   = FLA_Obj_length( A );
  inc_A    = 1;

  m_B      = FLA_Obj_length( B );
  n_B      = FLA_Obj_width( B );
  ldim_B   = FLA_Obj_length( B );
  inc_B    = 1;

  switch ( datatype ){

  case FLA_INT:
  case FLA_FLOAT:
  {
    float* buff_A_gpu = ( float* ) A_gpu;
    float* buff_B_gpu = ( float* ) B_gpu;

    for ( i = 0; i < n_B; i++ )
      cublasScopy( m_B,
                   buff_A_gpu + i * ldim_A, inc_A,
                   buff_B_gpu + i * ldim_B, inc_B );

    break;
  }

  case FLA_DOUBLE:
  {
    double* buff_A_gpu = ( double* ) A_gpu;
    double* buff_B_gpu = ( double* ) B_gpu;

    for ( i = 0; i < n_B; i++ )
      cublasDcopy( m_B,
                   buff_A_gpu + i * ldim_A, inc_A,
                   buff_B_gpu + i * ldim_B, inc_B );

    break;
  }

  case FLA_COMPLEX:
  {
    cuComplex* buff_A_gpu = ( cuComplex* ) A_gpu;
    cuComplex* buff_B_gpu = ( cuComplex* ) B_gpu;

    for ( i = 0; i < n_B; i++ )
      cublasCcopy( m_B,
                   buff_A_gpu + i * ldim_A, inc_A,
                   buff_B_gpu + i * ldim_B, inc_B );

    break;
  }

  case FLA_DOUBLE_COMPLEX:
  {
    cuDoubleComplex* buff_A_gpu = ( cuDoubleComplex* ) A_gpu;
    cuDoubleComplex* buff_B_gpu = ( cuDoubleComplex* ) B_gpu;

    for ( i = 0; i < n_B; i++ )
      cublasZcopy( m_B,
                   buff_A_gpu + i * ldim_A, inc_A,
                   buff_B_gpu + i * ldim_B, inc_B );

    break;
  }

  }
  
  return FLA_SUCCESS;
}