axpy.hpp

// FEAT3: Finite Element Analysis Toolbox, Version 3
// Copyright (C) 2010 by Stefan Turek & the FEAT group
// FEAT3 is released under the GNU General Public License version 3,
// see the file 'copyright.txt' in the top level directory for details.
 
#pragma once
#ifndef KERNEL_LAFEM_ARCH_AXPY_HPP
#define KERNEL_LAFEM_ARCH_AXPY_HPP 1
 
// includes, FEAT
#include <kernel/base_header.hpp>
#include <kernel/backend.hpp>
#include <kernel/util/half.hpp>
 
namespace FEAT
{
  namespace LAFEM
  {
    namespace Arch
    {
      struct Axpy
      {
        template <typename DT_>
        static void value(DT_ * r, const DT_ a, const DT_ * const x, const Index size)
        {
          value_generic(r, a, x, size);
        }
 
        template <typename ValueType_>
        static void value_blocked(ValueType_ * r, const ValueType_ a, const ValueType_ * const x, const Index size)
        {
          value_blocked_generic(r, a, x, size);
        }
 
#ifdef FEAT_HAVE_HALFMATH
        static void value(Half * r, const Half a, const Half * const x, const Index size)
        {
          BACKEND_SKELETON_VOID(value_cuda, value_generic, value_generic, r, a, x, size)
        }
#endif
 
        static void value(float * r, const float a, const float * const x, const Index size)
        {
          BACKEND_SKELETON_VOID(value_cuda, value_mkl, value_generic, r, a, x, size)
        }
 
        static void value(double * r, const double a, const double * const x, const Index size)
        {
          BACKEND_SKELETON_VOID(value_cuda, value_mkl, value_generic, r, a, x, size)
        }
 
        template <typename DT_>
        static void value_generic(DT_ * r, const DT_ a, const DT_ * const x, const Index size);
 
        static void value_mkl(float * r, const float a, const float * const x, const Index size);
        static void value_mkl(double * r, const double a, const double * const x, const Index size);
 
        template <typename DT_>
        static void value_cuda(DT_ * r, const DT_ a, const DT_ * const x, const Index size);
 
        template <typename ValueType_>
        static void value_blocked_generic(ValueType_ * r, const ValueType_ a, const ValueType_ * const x, const Index size);
      };
 
#ifdef FEAT_EICKT
      extern template void Axpy::value_generic(float *, const float, const float * const, const Index);
      extern template void Axpy::value_generic(double *, const double, const double * const, const Index);
#endif
 
    } // namespace Arch
  } // namespace LAFEM
} // namespace FEAT
 
#ifndef  __CUDACC__
#include <kernel/lafem/arch/axpy_generic.hpp>
#endif
#endif // KERNEL_LAFEM_ARCH_AXPY_HPP