GatherRegDecl.h

// GatherRegDecl.h:
// ------------
//

#ifndef __IPSDKUTIL_INSTRUCTIONSET_GATHER_DETAIL_AVX512_GATHERREGDECL_H__
#define __IPSDKUTIL_INSTRUCTIONSET_GATHER_DETAIL_AVX512_GATHERREGDECL_H__

#include <IPSDKUtil/IPSDKUtilExports.h>
#include <IPSDKUtil/InstructionSet/Gather/detail/GatherReg.h>
#include <IPSDKUtil/InstructionSet/InstructionSetTypes.h>
#include <IPSDKUtil/InstructionSet/AVX512/Avx512Types.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 GatherReg<eInstructionSet::eIS_Avx512, 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 T* baseAddress,
        const ipUInt32* indexes,
        typename Avx512Type<T>::Type& out);

    static
    IPSDK_FORCEINLINE
    void
    act(const T* baseAddress,
        const typename Avx512Type<ipUInt32>::Type& indexes1,
        const typename Avx512Type<ipUInt32>::Type& indexes2,
        const typename Avx512Type<ipUInt32>::Type& indexes3,
        const typename Avx512Type<ipUInt32>::Type& indexes4,
        typename Avx512Type<T>::Type& out);
};

template <typename T>
struct GatherReg<eInstructionSet::eIS_Avx512, 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 T* baseAddress,
        const ipUInt32* indexes,
        typename Avx512Type<T>::Type& out);

    static
    IPSDK_FORCEINLINE
    void
    act(const T* baseAddress,
        const typename Avx512Type<ipUInt32>::Type& indexes1,
        const typename Avx512Type<ipUInt32>::Type& indexes2,
        typename Avx512Type<T>::Type& out);
};

template <typename T>
struct GatherReg<eInstructionSet::eIS_Avx512, 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 T* baseAddress,
        const ipUInt32* indexes,
        typename Avx512Type<T>::Type& out);

    static
    IPSDK_FORCEINLINE
    void
    act(const T* baseAddress, const typename Avx512Type<ipUInt32>::Type& indexes,
        typename Avx512Type<T>::Type& out);
};

template <>
struct GatherReg<eInstructionSet::eIS_Avx512, ipReal32>
{
    static
    IPSDK_FORCEINLINE
    void
    act(const ipReal32* baseAddress,
        const ipUInt32* indexes,
        Avx512Type<ipReal32>::Type& out);

    static
    IPSDK_FORCEINLINE
    void
    act(const ipReal32* baseAddress, const Avx512Type<ipUInt32>::Type& indexes,
        Avx512Type<ipReal32>::Type& out);
};

template <>
struct GatherReg<eInstructionSet::eIS_Avx512, ipReal64>
{
    static
    IPSDK_FORCEINLINE
    void
    act(const ipReal64* baseAddress,
        const ipUInt32* indexes,
        Avx512Type<ipReal64>::Type& out);

    static
    IPSDK_FORCEINLINE
    void
    act(const ipReal64* baseAddress, const __m256i& indexes,
        Avx512Type<ipReal64>::Type& out);
};


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

#endif // __IPSDKUTIL_INSTRUCTIONSET_GATHER_DETAIL_AVX512_GATHERREGDECL_H__