Function transform_32f

deps/opencv/modules/core/src/matmul.cpp:1561–1616

static void
transform_32f( const float* src, float* dst, const float* m, int len, int scn, int dcn )
{
#if CV_SSE2
    if( USE_SSE2 )
    {
        int x = 0;
        if( scn == 3 && dcn == 3 )
        {
            __m128 m0, m1, m2, m3;
            load3x3Matrix(m, m0, m1, m2, m3);

            for( ; x < (len - 1)*3; x += 3 )
            {
                __m128 x0 = _mm_loadu_ps(src + x);
                __m128 y0 = _mm_add_ps(_mm_add_ps(_mm_add_ps(
                            _mm_mul_ps(m0, _mm_shuffle_ps(x0,x0,_MM_SHUFFLE(0,0,0,0))),
                            _mm_mul_ps(m1, _mm_shuffle_ps(x0,x0,_MM_SHUFFLE(1,1,1,1)))),
                            _mm_mul_ps(m2, _mm_shuffle_ps(x0,x0,_MM_SHUFFLE(2,2,2,2)))), m3);
                _mm_storel_pi((__m64*)(dst + x), y0);
                _mm_store_ss(dst + x + 2, _mm_movehl_ps(y0,y0));
            }

            for( ; x < len*3; x += 3 )
            {
                float v0 = src[x], v1 = src[x+1], v2 = src[x+2];
                float t0 = saturate_cast<float>(m[0]*v0 + m[1]*v1 + m[2]*v2 + m[3]);
                float t1 = saturate_cast<float>(m[4]*v0 + m[5]*v1 + m[6]*v2 + m[7]);
                float t2 = saturate_cast<float>(m[8]*v0 + m[9]*v1 + m[10]*v2 + m[11]);
                dst[x] = t0; dst[x+1] = t1; dst[x+2] = t2;
            }
            return;
        }

        if( scn == 4 && dcn == 4 )
        {
            __m128 m0, m1, m2, m3, m4;
            load4x4Matrix(m, m0, m1, m2, m3, m4);

            for( ; x < len*4; x += 4 )
            {
                __m128 x0 = _mm_loadu_ps(src + x);
                __m128 y0 = _mm_add_ps(_mm_add_ps(_mm_add_ps(_mm_add_ps(
                                    _mm_mul_ps(m0, _mm_shuffle_ps(x0,x0,_MM_SHUFFLE(0,0,0,0))),
                                    _mm_mul_ps(m1, _mm_shuffle_ps(x0,x0,_MM_SHUFFLE(1,1,1,1)))),
                                    _mm_mul_ps(m2, _mm_shuffle_ps(x0,x0,_MM_SHUFFLE(2,2,2,2)))),
                                    _mm_mul_ps(m3, _mm_shuffle_ps(x0,x0,_MM_SHUFFLE(3,3,3,3)))), m4);
                _mm_storeu_ps(dst + x, y0);
            }
            return;
        }
    }
#endif

    transform_(src, dst, m, len, scn, dcn);
}