float ssei[16];

    __m128 *pivot = (__m128*) ssei;
    __m128 *local = (__m128*) &ssei[4];
    __m128 *diff = (__m128*) &ssei[8];
    __m128 *rmul = (__m128*) &ssei[12];

    for(int i=0; i<16; i++)
        ssei[i] = 0.0f;



#ifdef USE_OPENMP
#pragma omp parallel for
#endif
    for (int VOut = totalSizeV - 1; VOut > endVOut; VOut--)
    {

        for (int HOut = beginHOut; HOut < endHOut; HOut++)
        {

            ssei[0] = data[VOut * totalSizeH + HOut];//
            ssei[1] = data[(VOut * totalSizeH + HOut) + globalSize];//
            ssei[2] = data[(VOut * totalSizeH + HOut) + 2 * globalSize];//

            int rowSize = (int) (100.0f / ssei[2]);//
            if (rowSize > 128)
                rowSize = 128;
            if (rowSize < 11)
                rowSize = 11;

            localSizeV = localSizeH = rowSize;

            int partSize = rowSize / 2;

            for (int VIn = VOut - localSizeV; VIn <= VOut; VIn++)
            {
                for (int HIn = HOut - partSize; HIn < partSize + HOut; HIn++)
                {
                    ssei[4] = data[VIn * totalSizeH + HIn];//
                    ssei[5] = data[(VIn * totalSizeH + HIn) + globalSize];//
                    ssei[6] = data[(VIn * totalSizeH + HIn) + 2 * globalSize];//

                    ssei[9] = ssei[5] - ssei[1];//
                    //*diff = _mm_sub_ps(local, pivot);

                    if (ssei[9] > hmin && ssei[9] < hmax)
                    {

                        ssei[8] = ssei[4] - ssei[0];//
                        ssei[10] = ssei[6] - ssei[2];//
                        //*rmul = _mm_mul_ps(*diff, *diff);

                        float norm = fastSqrt_Q3(ssei[8]*ssei[8] + ssei[9]*ssei[9] + ssei[10]*ssei[10]);//

                        float cosAlpha = ssei[9]/ norm;//

                        if (cosAlpha > cosalpha)
                            img[VIn * totalSizeH + HIn] = 1;
                    }
                }
            }
        }
    }