static void test_mul32s(void)
{
    int x, y;
    for(y = 0; y < height; y++)
    {
        uint32_t *data = img + y * width;

        for(x = 0; x < width ; x++)
        {
            uint32_t x = *data;
            uint8_t a = x >> 24;
            uint8_t b = (uint8_t)(x >> 16);
            uint8_t c = (uint8_t)(x >> 8);
            uint8_t d = (uint8_t)x;
            a *= K;
            b *= K;
            c *= K;
            d *= K;
            *data++ = ((uint32_t)a << 24) | ((uint32_t)b << 16)
                       | ((uint16_t)c << 8) | d;
        }
    }
}

static void test_mul32(void)
{
    int x, y;
    for(y = 0; y < height; y++)
    {
        uint32_t *data = img + y * width;

        for(x = 0; x < width ; x++)
        {
            union { uint32_t x; uint8_t t[4]; } u;
            u.x = *data;
            u.t[0] *= K;
            u.t[1] *= K;
            u.t[2] *= K;
            u.t[3] *= K;
            *data++ = u.x;
        }
    }
}