ScatterRegDecl.h

// ScatterRegDecl.h:
// ------------
//

#ifndef __IPSDKUTIL_INSTRUCTIONSET_SCATTER_DETAIL_SSE2_SCATTERREGDECL_H__
#define __IPSDKUTIL_INSTRUCTIONSET_SCATTER_DETAIL_SSE2_SCATTERREGDECL_H__

#include <IPSDKUtil/IPSDKUtilExports.h>
#include <IPSDKUtil/InstructionSet/Scatter/detail/ScatterReg.h>
#include <IPSDKUtil/InstructionSet/InstructionSetTypes.h>
#include <IPSDKUtil/InstructionSet/SSE2/Sse2Types.h>
#include <IPSDKUtil/Tools/ForceInline.h>

#include <boost/mpl/and.hpp>
#include <boost/mpl/equal_to.hpp>
#include <boost/mpl/int.hpp>
#include <boost/type_traits/is_same.hpp>
#include <boost/type_traits/is_signed.hpp>

namespace ipsdk {
namespace simd {
namespace detail {


template <typename T>
struct ScatterReg<eInstructionSet::eIS_Sse2, T,
    typename boost::enable_if<
        typename boost::mpl::and_<typename boost::is_integral<T>::type, 
                                  typename boost::mpl::equal_to<boost::mpl::int_<sizeof(T)>, boost::mpl::int_<1> >::type
        >::type
    >::type
>
{
    static
    IPSDK_FORCEINLINE
    void
    act(const typename Sse2Type<T>::Type& in,
        const ipUInt32* indexes,
        T* const outBaseAddr);

    static
    IPSDK_FORCEINLINE
    void
    act(const typename Sse2Type<T>::Type& in,
        const typename Sse2Type<ipUInt32>::Type& indexes1,
        const typename Sse2Type<ipUInt32>::Type& indexes2,
        const typename Sse2Type<ipUInt32>::Type& indexes3,
        const typename Sse2Type<ipUInt32>::Type& indexes4,
        T* const outBaseAddr);
};

template <typename T>
struct ScatterReg<eInstructionSet::eIS_Sse2, T,
    typename boost::enable_if<
        typename boost::mpl::and_<typename boost::is_integral<T>::type, 
                                  typename boost::mpl::equal_to<boost::mpl::int_<sizeof(T)>, boost::mpl::int_<2> >::type
        >::type
    >::type
>
{
    static
    IPSDK_FORCEINLINE
    void
    act(const typename Sse2Type<T>::Type& in,
        const ipUInt32* indexes,
        T* const outBaseAddr);

    static
    IPSDK_FORCEINLINE
    void
    act(const typename Sse2Type<T>::Type& in,
        const typename Sse2Type<ipUInt32>::Type& indexes1,
        const typename Sse2Type<ipUInt32>::Type& indexes2,
        T* const outBaseAddr);
};

template <typename T>
struct ScatterReg<eInstructionSet::eIS_Sse2, T,
    typename boost::enable_if<
        typename boost::mpl::and_<typename boost::is_integral<T>::type, 
                                  typename boost::mpl::equal_to<boost::mpl::int_<sizeof(T)>, boost::mpl::int_<4> >::type
        >::type
    >::type
>
{
    static
    IPSDK_FORCEINLINE
    void
    act(const typename Sse2Type<T>::Type& in,
        const ipUInt32* indexes,
        T* const outBaseAddr);

    static
    IPSDK_FORCEINLINE
    void
    act(const typename Sse2Type<T>::Type& in,
        const typename Sse2Type<ipUInt32>::Type& indexes,
        T* const outBaseAddr);
};

template <>
struct ScatterReg<eInstructionSet::eIS_Sse2, ipReal32>
{
    static
    IPSDK_FORCEINLINE
    void
    act(const Sse2Type<ipReal32>::Type& in,
        const ipUInt32* indexes,
        ipReal32* const outBaseAddr);

    static
    IPSDK_FORCEINLINE
    void
    act(const Sse2Type<ipReal32>::Type& in,
        const Sse2Type<ipUInt32>::Type& indexes,
        ipReal32* const outBaseAddr);
};

template <>
struct ScatterReg<eInstructionSet::eIS_Sse2, ipReal64>
{
    static
    IPSDK_FORCEINLINE
    void
    act(const Sse2Type<ipReal64>::Type& in,
        const ipUInt32* indexes,
        ipReal64* const outBaseAddr);
};


} // end of namespace detail
} // end of namespace simd
} // end of namespace ipsdk

#endif // __IPSDKUTIL_INSTRUCTIONSET_SCATTER_DETAIL_SSE2_SCATTERREGDECL_H__