AssignRegImpl.h

Go to the documentation of this file.

// AssignRegImpl.h:
// ------------
//

#ifndef __IPSDKUTIL_INSTRUCTIONSET_DETAIL_SSE2_ASSIGNREGIMPL_H__
#define __IPSDKUTIL_INSTRUCTIONSET_DETAIL_SSE2_ASSIGNREGIMPL_H__

#include <IPSDKUtil/InstructionSet/detail/SSE2/AssignRegDecl.h>

namespace ipsdk {
namespace simd {
namespace detail {


template <typename T>
IPSDK_FORCEINLINE
typename Sse2Type<T>::Type
AssignReg<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
>
::act(const T& value)
{
    return _mm_set1_epi8(value);
}

template <typename T>
IPSDK_FORCEINLINE
void
AssignReg<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
>
::act(typename Sse2Type<T>::Type& reg, const T& value)
{
    reg = _mm_set1_epi8(value);
}

template <typename T>
IPSDK_FORCEINLINE
typename Sse2Type<T>::Type
AssignReg<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
>::act(const T& value)
{
    return _mm_set1_epi16(value);
}

template <typename T>
IPSDK_FORCEINLINE
void
AssignReg<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
>::act(typename Sse2Type<T>::Type& reg, const T& value)
{
    reg = _mm_set1_epi16(value);
}

template <typename T>
IPSDK_FORCEINLINE
typename Sse2Type<T>::Type
AssignReg<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
>::act(const T& value)
{
    return _mm_set1_epi32(value);
}

template <typename T>
IPSDK_FORCEINLINE
void
AssignReg<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
>::act(typename Sse2Type<T>::Type& reg, const T& value)
{
    reg = _mm_set1_epi32(value);
}

template <typename T>
IPSDK_FORCEINLINE
typename Sse2Type<T>::Type
AssignReg<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_<8> >::type
        >::type
    >::type
>::act(const T& value)
{
    // there's no instruction such as "_mm_set1_epi32" for 64 bits integers,
    // so we use a work-around...
    T values[2];
    values[0] = values[1] = value;
    return _mm_loadu_si128(
        reinterpret_cast<const typename Sse2Type<T>::Type*>(values));
}

template <typename T>
IPSDK_FORCEINLINE
void
AssignReg<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_<8> >::type
        >::type
    >::type
>::act(typename Sse2Type<T>::Type& reg, const T& value)
{
    // there's no instruction such as "_mm_set1_epi32" for 64 bits integers,
    // so we use a work-around...
    T values[2];
    values[0] = values[1] = value;
    reg = _mm_loadu_si128(
        reinterpret_cast<const typename Sse2Type<T>::Type*>(values));
}

// assign implementation for SSE2 for real32 type
IPSDK_FORCEINLINE
Sse2Type<ipReal32>::Type
AssignReg<eInstructionSet::eIS_Sse2, ipReal32>::act(const ipReal32& value)
{
    return _mm_set1_ps(value);
}

IPSDK_FORCEINLINE
void
AssignReg<eInstructionSet::eIS_Sse2, ipReal32>::act(Sse2Type<ipReal32>::Type& reg,
                                                    const ipReal32& value)
{
    reg = _mm_set1_ps(value);
}

// assign implementation for SSE2 for real64 type
IPSDK_FORCEINLINE
Sse2Type<ipReal64>::Type
AssignReg<eInstructionSet::eIS_Sse2, ipReal64>::act(const ipReal64& value)
{
    return _mm_set1_pd(value);
}

IPSDK_FORCEINLINE
void
AssignReg<eInstructionSet::eIS_Sse2, ipReal64>::act(Sse2Type<ipReal64>::Type& reg,
                                                    const ipReal64& value)
{
    reg = _mm_set1_pd(value);
}


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

#endif // __IPSDKUTIL_INSTRUCTIONSET_DETAIL_SSE2_ASSIGNREGIMPL_H__