MaxReg.h

Go to the documentation of this file.

// MaxReg.h:
// -------------------
//

#ifndef __IPSDKUTIL_INSTRUCTIONSET_REDUCTION_DETAIL_AVX_MAXREG_H__
#define __IPSDKUTIL_INSTRUCTIONSET_REDUCTION_DETAIL_AVX_MAXREG_H__

#include <IPSDKUtil/IPSDKUtilExports.h>
#include <IPSDKUtil/InstructionSet/cast.h>
#include <IPSDKUtil/Tools/ForceInline.h>

namespace ipsdk {
namespace simd {
namespace detail {
namespace reduction {


template <>
struct MaxReg<eInstructionSet::eIS_Avx, ipReal32>
{
    static IPSDK_FORCEINLINE
    ipReal32
    act(const AvxType<ipReal32>::Type& in)
    {
        ipReal32 unloaded[8];
        UnloadReg<eInstructionSet::eIS_Avx, ipReal32>::act(in, unloaded);
        ipReal32 tmax = unloaded[0];
        for(int i=1; i<8; ++i) {
            tmax = (unloaded[i] > tmax ? unloaded[i] : tmax);
        }
        return tmax;
    }
};

template <>
struct MaxReg<eInstructionSet::eIS_Avx, ipReal64>
{
    static IPSDK_FORCEINLINE
    ipReal64
    act(const AvxType<ipReal64>::Type& in)
    {
        ipReal64 unloaded[4];
        UnloadReg<eInstructionSet::eIS_Avx, ipReal64>::act(in, unloaded);
        ipReal64 tmax = unloaded[0];
        for(int i=1; i<4; ++i) {
            tmax = (unloaded[i] > tmax ? unloaded[i] : tmax);
        }
        return tmax;
    }
};


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

#endif // __IPSDKUTIL_INSTRUCTIONSET_REDUCTION_DETAIL_AVX_MAXREGIMPL_H__