ScatterPackImpl.h

Go to the documentation of this file.

// ScatterPackImpl.h:
// ------------
//

#ifndef __IPSDKUTIL_INSTRUCTIONSET_DETAIL_AVX2_SCATTERPACKIMPL_H__
#define __IPSDKUTIL_INSTRUCTIONSET_DETAIL_AVX2_SCATTERPACKIMPL_H__

#include <IPSDKUtil/IPSDKUtilExports.h>
#include <IPSDKUtil/InstructionSet/Scatter/detail/ScatterPackImpl.h>
#include <IPSDKUtil/InstructionSet/Scatter/detail/ScatterReg.h>
#include <IPSDKUtil/InstructionSet/detail/Std/BasePack.h>
#include <IPSDKUtil/Tools/ForceInline.h>

namespace ipsdk {
namespace simd {
namespace detail {


template <typename T>
IPSDK_FORCEINLINE
void
ScatterPack<eInstructionSet::eIS_Avx2, T,
    typename boost::enable_if_c<sizeof(T) == 1>::type
>::act(
    const BasePack<ePackType::ePT_Avx, T>& in,
    const ipUInt32* indexes,
    T* const outBaseAddr)
{
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[0],
        indexes,
        outBaseAddr);
}

template <typename T>
IPSDK_FORCEINLINE
void
ScatterPack<eInstructionSet::eIS_Avx2, T,
    typename boost::enable_if_c<sizeof(T) == 1>::type
>::act(
    const BasePack<ePackType::ePT_Avx, T>& in,
    const BasePack<ePackType::ePT_Avx, ipUInt32>& packIdx,
    T* const outBaseAddr)
{
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[0],
        packIdx._val[0], packIdx._val[1],
        packIdx._val[2], packIdx._val[3],
        outBaseAddr);
}

template <typename T>
IPSDK_FORCEINLINE
void
ScatterPack<eInstructionSet::eIS_Avx2, T,
    typename boost::enable_if_c<sizeof(T) == 2>::type
>::act(
    const BasePack<ePackType::ePT_Avx, T>& in,
    const ipUInt32* indexes,
    T* const outBaseAddr)
{
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[0], indexes, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[1], indexes+16, outBaseAddr);
}

template <typename T>
IPSDK_FORCEINLINE
void
ScatterPack<eInstructionSet::eIS_Avx2, T,
    typename boost::enable_if_c<sizeof(T) == 2>::type
>::act(
    const BasePack<ePackType::ePT_Avx, T>& in,
    const BasePack<ePackType::ePT_Avx, ipUInt32>& packIdx,
    T* const outBaseAddr)
{
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[0], packIdx._val[0], packIdx._val[1], outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[1], packIdx._val[2], packIdx._val[3], outBaseAddr);
}

template <typename T>
IPSDK_FORCEINLINE
void
ScatterPack<eInstructionSet::eIS_Avx2, T,
    typename boost::enable_if_c<sizeof(T) == 4>::type
>::act(
    const BasePack<ePackType::ePT_Avx, T>& in,
    const ipUInt32* indexes,
    T* const outBaseAddr)
{
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[0], indexes, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[1], indexes+8, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[2], indexes+16, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[3], indexes+24, outBaseAddr);
}

template <typename T>
IPSDK_FORCEINLINE
void
ScatterPack<eInstructionSet::eIS_Avx2, T,
    typename boost::enable_if_c<sizeof(T) == 4>::type
>::act(
    const BasePack<ePackType::ePT_Avx, T>& in,
    const BasePack<ePackType::ePT_Avx, ipUInt32>& packIdx,
    T* const outBaseAddr)
{
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[0], packIdx._val[0], outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[1], packIdx._val[1], outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[2], packIdx._val[2], outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[3], packIdx._val[3], outBaseAddr);
}

template <typename T>
IPSDK_FORCEINLINE
void
ScatterPack<eInstructionSet::eIS_Avx2, T,
    typename boost::enable_if_c<sizeof(T) == 8>::type
>::act(
    const BasePack<ePackType::ePT_Avx, T>& in,
    const ipUInt32* indexes,
    T* const outBaseAddr)
{
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[0], indexes, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[1], indexes+4, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[2], indexes+8, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[3], indexes+12, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[4], indexes+16, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[5], indexes+20, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[6], indexes+24, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[7], indexes+28, outBaseAddr);
}

template <typename T>
IPSDK_FORCEINLINE
void
ScatterPack<eInstructionSet::eIS_Avx2, T,
    typename boost::enable_if_c<sizeof(T) == 8>::type
>::act(
    const BasePack<ePackType::ePT_Avx, T>& in,
    const BasePack<ePackType::ePT_Avx, ipUInt32>& packIdx,
    T* const outBaseAddr)
{
    ipUInt32 indexes[32];
    UnloadPack<eInstructionSet::eIS_Avx2, ipUInt32>::act(packIdx, indexes);
    
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[0], indexes, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[1], indexes+4, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[2], indexes+8, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[3], indexes+12, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[4], indexes+16, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[5], indexes+20, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[6], indexes+24, outBaseAddr);
    ScatterReg<eInstructionSet::eIS_Avx2, T>::act(
        in._val[7], indexes+28, outBaseAddr);
}


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

#endif // __IPSDKUTIL_INSTRUCTIONSET_SCATTER_DETAIL_AVX2_SCATTERPACKIMPL_H__