Opencv中goodFeaturesToTrack函數(Harris角點、Shi-Tomasi角點檢測)運算元速度的進一步優化（1920*1080測試圖11ms處理完成）。

void cv::goodFeaturesToTrack( InputArray _image, OutputArray _corners,
                              int maxCorners, double qualityLevel, double minDistance,
                              InputArray _mask, int blockSize,
                              bool useHarrisDetector, double harrisK )
{
    Mat image = _image.getMat(), eig, tmp;
    if( useHarrisDetector )
        cornerHarris( image, eig, blockSize, 3, harrisK );
    else
        cornerMinEigenVal( image, eig, blockSize, 3 );
    double maxVal = 0;
    minMaxLoc( eig, 0, &maxVal, 0, 0, _mask );
    threshold( eig, eig, maxVal*qualityLevel, 0, THRESH_TOZERO );
    dilate( eig, tmp, Mat());
    Size imgsize = image.size();
    std::vector<const float*> tmpCorners;
    // collect list of pointers to features - put them into temporary image
    Mat mask = _mask.getMat();
    for( int y = 1; y < imgsize.height - 1; y++ )
    {
        const float* eig_data = (const float*)eig.ptr(y);
        const float* tmp_data = (const float*)tmp.ptr(y);
        const uchar* mask_data = mask.data ? mask.ptr(y) : 0;
        for( int x = 1; x < imgsize.width - 1; x++ )
        {
            float val = eig_data[x];
            if( val != 0 && val == tmp_data[x] && (!mask_data || mask_data[x]) )
                tmpCorners.push_back(eig_data + x);
        }
    }
    std::sort( tmpCorners.begin(), tmpCorners.end(), greaterThanPtr() );
    std::vector<Point2f> corners;
    size_t i, j, total = tmpCorners.size(), ncorners = 0;
    if (minDistance >= 1)
    {
         // Partition the image into larger grids
    }
    else
    {
       
    }
    Mat(corners).convertTo(_corners, _corners.fixedType() ? _corners.type() : CV_32F);
}

if( val != 0 && val == tmp_data[x] && (!mask_data || mask_data[x]) )

            __m128i DxDy_L = _mm_mullo_epi32(_mm_cvtepi16_epi32(Dx), _mm_cvtepi16_epi32(Dy));
            __m128i DxDy_H = _mm_mullo_epi32(_mm_cvtepi16_epi32(_mm_srli_si128(Dx, 8)), _mm_cvtepi16_epi32(_mm_srli_si128(Dy, 8)));

            __m128i L = _mm_mullo_epi16(Dx, Dy);
            __m128i H = _mm_mulhi_epi16(Dx, Dy);
            __m128i DxDy_L = _mm_unpacklo_epi16(L, H);
            __m128i DxDy_H = _mm_unpackhi_epi16(L, H);

int IM_FastestBoxBlur3X3_I16(short *Src, short *Dest, int Width, int Height)
{
    if ((Src == NULL) || (Dest == NULL))                    return IM_STATUS_NULLREFRENCE;
    if ((Width <= 0) || (Height <= 0))                        return IM_STATUS_INVALIDPARAMETER;

    short *RowCopy = (short *)malloc((Width + 2) * 3 * sizeof(short));
    if (RowCopy == NULL)    return IM_STATUS_OUTOFMEMORY;

    short *First = RowCopy;
    short *Second = RowCopy + (Width + 2);
    short *Third = RowCopy + (Width + 2) * 2;

    memcpy(Second, Src, sizeof(short));
    memcpy(Second + 1, Src, Width * sizeof(short));                                                        //    拷貝資料到中間位置
    memcpy(Second + (Width + 1), Src + (Width - 1), sizeof(short));

    memcpy(First, Second, (Width + 2) * sizeof(short));                                                    //    第一行和第二行一樣

    memcpy(Third, Src + Width, sizeof(short));                                                            //    拷貝第二行資料
    memcpy(Third + 1, Src + Width, Width * sizeof(short));
    memcpy(Third + (Width + 1), Src + Width + (Width - 1), sizeof(short));

    int BlockSize = 8, Block = Width / BlockSize;            //    測試表面一次性處理4個畫素處理8個要快一些

    for (int Y = 0; Y < Height; Y++)
    {
        short *LinePS = Src + Y * Width;
        short *LinePD = Dest + Y * Width;
        if (Y != 0)
        {
            short *Temp = First; First = Second; Second = Third; Third = Temp;
        }
        if (Y == Height - 1)
        {
            memcpy(Third, Second, (Width + 2) * sizeof(short));
        }
        else
        {
            memcpy(Third, Src + (Y + 1) * Width, sizeof(short));
            memcpy(Third + 1, Src + (Y + 1) * Width, Width * sizeof(short));                            //    由於備份了前面一行的資料，這裡即使Src和Dest相同也是沒有問題的
            memcpy(Third + (Width + 1), Src + (Y + 1) * Width + (Width - 1), sizeof(short));
        }
        for (int X = 0; X < Block * BlockSize; X += BlockSize)
        {
            __m128i P0 = _mm_loadu_si128((__m128i *)(First + X));
            __m128i P1 = _mm_loadu_si128((__m128i *)(First + X + 1));
            __m128i P2 = _mm_loadu_si128((__m128i *)(First + X + 2));

            __m128i P3 = _mm_loadu_si128((__m128i *)(Second + X));
            __m128i P4 = _mm_loadu_si128((__m128i *)(Second + X + 1));
            __m128i P5 = _mm_loadu_si128((__m128i *)(Second + X + 2));

            __m128i P6 = _mm_loadu_si128((__m128i *)(Third + X));
            __m128i P7 = _mm_loadu_si128((__m128i *)(Third + X + 1));
            __m128i P8 = _mm_loadu_si128((__m128i *)(Third + X + 2));

            __m128i Sum0123 = _mm_adds_epi16(_mm_adds_epi16(P0, P1), _mm_adds_epi16(P2, P3));
            __m128i Sum5678 = _mm_adds_epi16(_mm_adds_epi16(P5, P6), _mm_adds_epi16(P7, P8));
            __m128i Sum = _mm_adds_epi16(_mm_adds_epi16(Sum0123, Sum5678), P4);

            _mm_storeu_si128((__m128i *)(LinePD + X), Sum);
        }
        for (int X = Block * BlockSize; X < Width; X++)
        {
            int P0 = First[X], P1 = First[X + 1], P2 = First[X + 2];
            int P3 = Second[X], P4 = Second[X + 1], P5 = Second[X + 2];
            int P6 = Third[X], P7 = Third[X + 1], P8 = Third[X + 2];
            LinePD[X] = P0 + P1 + P2 + P3 + P4 + P5 + P6 + P7 + P8;
        }
    }
    free(RowCopy);
    return IM_STATUS_OK;
}